Agent-Based Model

A simulation approach representing individual entities, such as patients or providers, as autonomous agents following behavioural rules, letting system patterns emerge.

Source: Bonabeau 2002

An agent-based model is a simulation approach that represents individual entities, such as patients or providers, as autonomous agents following behavioural rules, and lets patterns at the level of the whole system emerge from their interactions. Rather than specifying system behaviour directly, it specifies how individual agents act and interact, and observes what results. Bonabeau described how this bottom-up approach can capture phenomena, such as the spread of behaviour or disease, that arise from many individual decisions.

Source: Bonabeau 2002

An agent-based model is a simulation built from many individual agents, each following its own rules and interacting with others and with a shared environment. System-level patterns emerge from these local interactions rather than being specified directly. Agents can represent people, organisations or other entities, and they may adapt their behaviour in response to what happens around them. This bottom-up structure suits problems where heterogeneity, networks or feedback between individuals matter. In health, it has been applied to infectious disease spread and to the behaviour of care systems.

Source: Marshall et al. 2015

An agent-based model works by defining a population of agents, each with characteristics and rules governing how it behaves and responds to others and its environment, then simulating their interactions over time. System-level outcomes, such as epidemics, market patterns, or congestion, emerge from the accumulated actions of the agents rather than being imposed. By running the simulation, the modeller observes how individual behaviour and interaction give rise to aggregate patterns, which can be explored under different rules and conditions.

Source: Bonabeau 2002

An agent-based model represents individuals separately, each acting by its own rules, and derives system behaviour from their interactions, whereas an aggregate model represents the population by summary parameters, such as average rates, without distinguishing individuals. Agent-based models can capture heterogeneity, interaction, and emergent effects that aggregate models miss, but they are more demanding in data and computation. Aggregate models are simpler and often sufficient when individual variation and interaction do not much affect the outcome of interest.

Source: Bonabeau 2002

An agent-based model is useful in health where outcomes depend on interactions between individuals or on heterogeneity that aggregate models cannot capture, such as the spread of infectious disease through contact, the diffusion of behaviours, or the effects of individual decisions in a health system. By modelling agents and their interactions, it can represent transmission, networks, and adaptation that population-average models omit. It suits questions where the emergent consequences of many individual actions are central to the outcome.

Source: Bonabeau 2002

Agent-based models are demanding in data, since they require specifying the behaviour and interactions of individual agents, which may not be well known, and in computation, since simulating many agents is intensive. Their results can be sensitive to the rules and assumptions chosen, which are often uncertain, and the complexity can make them hard to validate and to interpret. The flexibility that lets them capture emergence also risks over-fitting or unwarranted confidence, so their assumptions and behaviour must be examined carefully.

Source: Bonabeau 2002