This is an excellent example of using an agent-based model (ABM) to ask “what if” questions about aspects of individual behaviour that would be very difficult to replicate within a real population. In this case the positioning of different types of food outlets within neighbourhoods.

This is a model of the inequalities in diet in the context of urban residential segregation. Auchincloss and Garcia use this model as an example of how to create an ABM to address complex non-communicable disease (NCD) problems. The model is used as an introductory guide to the philosophy and practices of ABMs and is used to illustrate why ABMs are a good fit for investigating NCD prevention. The paper describes very clearly how to take descriptions of individual behaviour and incorporate them into an ABM framework and run “what if “ experiments with the model.

This model investigates the inequalities in diet in the context of urban residential segregation

One of the contributory factors to obesity and obesity-related health conditions and illness is the high consumption of nutrient-poor energy-dense foods. Numerous population studies have shown that there is a higher prevalence of nutrient-poor energy-dense foods in the diets of lower income families compared with higher income families. One of the contributing factors to this difference in diet quality is access to healthy food outlets. It has long been recognised that urban landscapes are influenced by residential segregation based on both economic and ethnic factors and that there is a marked difference in resources and services between these areas. (See the model by Thomas Schelling included in this series for an ABM of segregation). Lower income neighbourhoods usually have far fewer outlets that provide healthy foods compared to higher income neighbourhoods. Auchincloss and Garcia’s model looks at the role of economic segregation on food outlet distribution and looks at possible “policy levers” that may be used to counteract this.

As is demonstrated by the Schelling model segregation and an individual's response to it is a function of the spatial location of populations. The nature of the spatial location of the individuals is governed by local factors and tends to be heterogeneous. By that we mean that the spatial pattern of the population tends not to a smooth gradation but tends to be a patchier pattern. This type of spatial distribution is very hard to incorporate in traditional mathematical or statistical models but is very easy to implement with an ABM. In this model we are mainly looking at the decisions that outlets make on their location, and food and price. The rule-based nature of ABMs makes construction of a shop based model both simpler and transparent.

Here we present a simplified version of Auchincloss’ model that concentrates on just the effect that segregation has on the success and type of food outlets within a neighbourhood

The model has two entities, people and shops. Each person is represented by a small dot. Each shop is represented by a larger circle. The colour of the shop circle indicates the type of shop: High end shop, Convenience shop.The colour of each person (agent) dot matches the dot colour in the centre of the shop circle that they are a customer of.

At the start of the simulation each person is allocated a square (a home) on the grid. A class is of neighbourhood is assigned to each home : affluent (indicated by a dark grey background) , non-affluent (light grey background). If the simulation is non-segregated then the neighbourhood class is assigned randomly. If it is a segregated simulation the affluent neighbourhood is assigned the top have of the grid.

At each iteration each person chooses the nearest shop. The shop they choose is indicated by the color of the dot of the person. Each person spends a set amount at the shop. A person from an affluent neighborhood will spend more at a high-end shop than a person from a non-affluent neighborhood. They both spend the same amount at the convenience store.

The number of shops can be set with the slider. At the start, there are equal numbers of high-end and convenience shops. The distribution of neighbourhoods (segregated or non-segregated) is set with the check box. You can rerun the model by stopping it and the clicking reset before running again.

At each iteration each person chooses the nearest shop. The shop they choose is indicated by the colour of the dot of the person. Each person spends a set amount at the shop. A person from an affluent neighbourhood will spend more at a high-end shop than a person from a non-affluent neighbourhood. They both spend the same amount at the convenience store.

At the end of an iteration the least profitable shop will relocate to a random spot in the grid and randomly assign its type to either high-end or convenience.

When the simulation is run with a segregated neighbourhood you can see that there are less stores in the non-affluent area and of those shops there are more convenience stores.

This just emerges out of the fact that affluent shoppers spend more in high-end shops. The convenience stores need a larger number of customers to make the same profit the thus creating food deserts in the non-affluent area

An interesting effect can be seen when the model is run with a non-segregated neighbourhood and a higher shop number. Here the number of convenience store decreases and are replaced by more stable high-end stores.

In a follow-up paper in which Auchincloss uses this model of residential segregation to look at the impact of income inequalities on diet in more detail, [2] the author concludes: "The model underscores the challenges of fostering favorable behavior change when people and resources are residentially segregated and behaviors are motivated or constrained by multiple factors. Simulation modeling can be a useful tool for proposing and testing policies or interventions that will ultimately be implemented in a complex system where the consequences of multidimensional interactions are diffıcult to predict."

[1]Auchincloss, Amy H., and Leandro Martin Totaro Garcia. "Brief introductory guide to agent-based modeling and an illustration from urban health research." Cadernos de saude publica 31 (2015): 65-78.

[2]Auchincloss, Amy H., et al. "An agent-based model of income inequalities in diet in the context of residential segregation." American journal of preventive medicine 40.3 (2011): 303-311.