Scientific models are essential for understanding and solving problems, particularly when we want to explore different pathways for the future. Just as we might build a model of an airplane by replicating the basic form, stripped of most details, at a 1:500 scale, we build a model of a system – in society, in nature, or at the intersection of the two – by creating a set of equations that distills the key elements and relationships within the system.

The problem is many of the resulting models are too simple. A common criticism of economic models, for example, is that many assume people will make rational decisions geared to maximizing economic gain – clearly not the case in real life. Another common problem with models is uniformity: they tend to look at society as a whole, ignoring the vast differences in knowledge, power, resources and priorities among different groups.

Agent-based modelling aims to do a better job of simulating the real world. Below, SEI Senior Research Fellow Richard Taylor explains what how ABM works and how SEI is beginning to apply it to address complex policy problems.

Q: What is agent-based modelling, and how is it different from conventional models?
RT: ABM is a type of modelling that describes the system at the level of the social actors within it – that is, the individual entities, each with their own goals, values, rules, information, knowledge, strategies and social context. In contrast, most models look at society at the macro-level, in aggregate, without distinguishing among different groups or entities.

In ABM, you identify the key actors (or groups) and describe them in some detail – looking at their values, their goals, their behaviours, and how they interact with one another. Often what we find is that even just setting a few rules at the micro-level can produce interesting changes at the macro-level. ABMs thus help us understand the links between micro and macro: for example, how the behaviours of different types of farmers affect deforestation rates.

Q: What kinds of situations are ABMs best suited for?
RT: ABMs are very flexible tools, but they take time to build and understand, because they include so much more detail than other models. They are particularly useful when there is a lot of heterogeneity in the system: to put it very simply, are you modelling a set of apples, or a set of apples, oranges, strawberries and bees? You don’t always know which details matter at the outset, so it is important to examine the components of your system to determine whether you need to disaggregate. But in real life, complexity and heterogeneity are increasingly the norm.

Q:How do you build an ABM?
RT: We implement the model in a computer program, but it is a good idea to start off with a “pen and paper” model. We start by writing down a set of linked issues and some ideas for who are the important social actors and how they interact. We have then to think about the boundaries of the model, what needs to be included and what can be left out. You need to think about what you want to find out with your model. Local information and participatory inquiry are invaluable in defining the research questions and developing the model.

Building an ABM is an exploratory process. The models can surprise you and expose uncertainties you were not expecting. You know all the variables that you have defined, but you don’t know how the simulations will behave until you test them out.

Q: How do you engage with stakeholders when building and applying ABMs?
RT: We have been using ABMs as a stakeholder engagement tool. They can have input at different stages of conceptualization, development and use of the model. People seem to find ABMs an engaging process, compared with other types of models, partly because of the detail of local information that can be included, I imagine. This can be visualized very nicely, and the models can be interactive, so that the users can configure the models themselves.

We have used ABMs in workshops with different stakeholders to get their views and refine the model and get these views represent the different levels of decision-making. So, it is one way to help bridge local resource user views and perceptions shared with policy-makers and potentially influence the thinking.

Q: What projects have you built ABMs for so far?
RT: We have worked on a range of things – for example, agriculture and forest-based livelihoods and the co-benefits of adaptation and mitigation to climate change in the Congo Basin.

Using ABM allowed us to include scientific data about land use and forest cover with information from residents about their farming practices, types of cropping, and other forest activities there. But with ABM one can explore different options. For example, this more detailed picture could also be used to understand how the situation may develop in the future, under scenarios including different forest policies, changing demographics and climate variation and change.

In a completely different context, looking at disaster resilience in Europe, I developed a model of the social and psychological factors in household disaster preparedness; that was part of the emBRACE project. Once we understood households’ motivations, we decided to include an insurance scenario that our model suggests would encourage preparedness.

Learn more on weADAPT:
https://www.weadapt.org/knowledge-base/adaptation-decision-making/identifying-salient-drivers-of-livelihood-decision-making-in-the-forest-communities-of-cameroon

https://www.weadapt.org/knowledge-base/adaptation-decision-making/agent-based-modelling

https://www.weadapt.org/knowledge-base/adaptation-decision-making/shrimp-aquaculture-in-bangladesh