2.1.1: What is ABM?#
In a nutshell
This section answers a single question: what exactly is an agent-based model, and why does getting the definition right matter? Pinning the definition down carefully pays off later, because every modelling choice you make follows from it. The section moves through five steps: the definition, the characteristics that set ABM apart, the purposes it can serve, two worked examples, and a short vocabulary recap.
2.1.1.1: Concept and definition of ABM#
A precise definition is the foundation everything else rests on. Getting it right matters because every later choice you make when building a model, from what an agent knows to how agents meet, follows from it, and a loose definition quietly lets the wrong assumptions creep in.
Agent-based modelling (ABM) is a computational approach for exploring and simulating the dynamic behaviour of complex systems composed of many interacting entities called agents[1]. Unlike equation-based models that aggregate behaviour into system-level variables, ABM takes a bottom-up perspective, representing each individual and its interactions explicitly.
Definition 1 (Agent-based model)
Agent-based models are computational simulation models that involve many discrete agents. (After Sayama, 2015[1].)
This simple definition carries three important ideas:
Agents’ rules are implemented algorithmically rather than expressed only as mathematical equations. This allows for rich behavioural rules and internal attributes (e.g., knowledge, emotion, memory) to be represented directly in code.
ABMs are typically used when studying systems made of numerous interacting agents (e.g., households, vehicles, firms). Their purpose is to capture collective, emergent behaviour that cannot be understood by studying individuals in isolation.
Each agent is an identifiable, autonomous entity with a clear boundary separating it from others and from its environment.
Because this discrete, bottom-up view is so general, it also connects to model families you may already know. Sayama (2015)[1] notes that ABM is a general framework that encompasses cellular automata and network models as special cases, because both represent systems made of discrete interacting units. As a result, ABMs have been applied across many domains.
The definition above describes what an ABM does; a complementary structural view describes what an ABM is made of. From this perspective, “the components of an agent-based model are a collection of agents and their states, the rules governing the interactions of the agents and the environment within which they live”[2].
At heart, an agent-based model is a way of describing a whole system as a crowd of separate individuals, each acting on its own, rather than as a single formula or a scene run by one controller. The activity below lets you build the definition one piece at a time, choosing the best phrase for each of four gaps in a single sentence.
The final piece of that activity names the idea that motivates the whole approach: emergence. It is worth stating clearly before we go further, because the rest of the section leans on it.
Definition 2 (Emergence)
Emergence is the appearance of system-wide patterns that arise from the many local interactions of individual agents, without anyone programming those patterns in directly. Traffic jams, market prices, and flocking are all emergent: they cannot be read off a single equation and only appear once the agents are set loose to interact. Chapter section 2.2.5 returns to emergence in full, including the micro-to-macro distinction.
2.1.1.2: Characteristics of ABM#
If the definition tells us what an ABM is, its defining characteristics tell us what sets it apart from other ways of modelling. Seven features, taken together, mark out the agent-based way of modelling. It works at the level of individuals rather than averages, builds system behaviour from the bottom up, keeps control decentralised, lets agents differ from one another and adapt as they go, treats each as a discrete identifiable entity, and places them in a space or network where position shapes who meets whom. These features matter because they decide what kinds of question a model can answer: a model that only tracks totals cannot show how one person’s choice ripples outward, whereas an agent-based model can. In the cards below, each of the seven characteristics is stated in brief. Click any card to turn it over and read a concrete example, and notice how each feature points to something an aggregate model would miss.
Important
Key insight: These characteristics combine to enable the study of complex adaptive systems, in which individual behaviours aggregate into emergent collective behaviour.
These characteristics make ABM particularly suitable for systems where individual heterogeneity, local interactions, adaptation, and spatial/network effects are important. Knowing what ABM is good at raises the next question: what do people actually build these models for?
2.1.1.3: Purposes of ABM#
ABM serves multiple purposes in research, policy, and education[3]. The tabs below present seven core purposes, each accompanied by practical guidance on its primary focus, the situations to which it is best suited, representative examples, and key considerations. This guidance supports the selection of an appropriate purpose for a given project. The final “Other” tab extends beyond these seven to a broader set of reasons for modelling.
Choosing the right purpose:
Focus on: Empirical validation and parameter calibration
Best for: Forecasting future scenarios, policy analysis, decision support
Examples: Epidemic spread, market dynamics, urban growth patterns
Key considerations:
Requires high-quality data for validation
Statistical analysis and uncertainty quantification
Model verification against real-world outcomes
Parameter sensitivity testing
Focus on: Mechanism representation and sensitivity analysis
Best for: Understanding why phenomena occur, theory development
Examples: Residential segregation (Schelling’s model), cooperation evolution, traffic congestion
Key considerations:
Detailed representation of causal mechanisms
Systematic exploration of parameter space
Clear documentation of assumptions
Focus on emergent patterns and their causes
Focus on: Detailed system representation capturing diversity and complexity
Best for: Comprehensive system documentation, baseline modelling
Examples: Ecosystem dynamics, organisational structures, supply chain networks
Key considerations:
Capture system heterogeneity and complexity
Document all relevant components and relationships
Balance detail with computational efficiency
Provide comprehensive system overview
Focus on: Testing and developing theories through experimentation
Best for: Academic research, hypothesis testing, theory validation
Examples: Social theories, economic market efficiency, evolutionary adaptation
Key considerations:
Rigorous experimental design
Systematic hypothesis testing
Clear theoretical framework
Reproducible results and methodology
Focus on: Clarity and visual communication
Best for: Teaching, demonstrations, public engagement
Examples: Market simulation games, predator-prey models, network effects
Key considerations:
Simple, understandable rules
Strong visual elements and animations
Interactive features for exploration
Clear explanation of model purpose and limitations
Focus on: Using simplified models as metaphors for complex systems
Best for: Cross-domain insights, conceptual understanding
Examples: Ant colonies for traffic optimisation, bird flocking for coordination
Key considerations:
Clear mapping between analogy and target system
Appropriate level of abstraction
Focus on key similarities and differences
Avoid over-interpretation of analogical relationships
Focus on: Stakeholder interaction and engagement
Best for: Participatory modelling, workshops, collaborative decision-making
Examples: Urban planning workshops, disaster preparedness, educational simulations
Key considerations:
Involve stakeholders in model development
Design for group interaction and discussion
Balance simplicity with realistic representation
Facilitate shared understanding and consensus building
Additional reasons for modelling and simulation:[4]
Support a decision: Models provide evidence and insights to inform decision-making processes
Help compare alternative options systematically
Quantify trade-offs and potential outcomes
Reduce uncertainty in complex decisions
Set an agenda: Models can highlight important issues and shape research or policy priorities
Identify critical factors that need attention
Frame problems in new ways
Guide resource allocation and planning
To learn how to model: The process of building models develops analytical and computational skills
Practice translating real-world problems into formal representations
Develop programming and mathematical modelling capabilities
Learn to think systematically about complex systems
Understanding ourselves: Models help us examine our own assumptions and mental models
Make implicit beliefs explicit and testable
Challenge our intuitions about how systems work
Reflect on our own cognitive biases and limitations
Combinations: Many modelling projects serve multiple purposes simultaneously, combining prediction with explanation, or illustration with decision support.
A model is never built for its own sake; it is built to serve a purpose, and naming that purpose shapes everything that follows. The same subject can be modelled to predict what will happen next, to explain why something occurs, to describe a system in detail, to develop a theory, to illustrate an idea, to draw an analogy, or to help a group of people learn together. Being clear about which of these you are after keeps the model honest, because it sets what counts as success and what the model can fairly be asked to do. In the activity below you are given a goal in each round. Read it, decide which purpose it serves, and watch how often the wording of a goal already tells you the answer.
The lesson to carry away is that stating the purpose of a model is the first design decision you make, and it quietly governs how the model should be built, judged, and trusted.
2.1.1.4: Examples of an ABM#
With the definition, characteristics, and purposes in hand, two worked examples make the ideas concrete. Both are chosen to show the same point: simple local rules, followed by many individual agents, give rise to coordinated large-scale patterns that no one designed directly, which is emergence in action.
Read each example at two levels. At the micro level, identify what state an agent carries, what local information it can observe, and which rule changes its state. At the macro level, identify the pattern measured across the population. The explanatory task of ABM is to connect those levels: changing a local rule or interaction should provide a reason for a corresponding change in the collective pattern, rather than merely producing an animation that looks plausible.
2.1.1.4.1: Traffic flow simulation#
To illustrate how ABM works in practice, consider a traffic flow model where individual cars (agents) move around a circular track. Each car follows simple rules:
Follow the car ahead: Maintain a safe distance from the vehicle in front
Adjust speed: Speed up when there’s space, slow down when approaching another car
Avoid collisions: React to nearby vehicles to prevent crashes
From these simple individual behaviours, complex traffic patterns emerge naturally:
Traffic jams: Areas where cars bunch up and move slowly
Traffic waves: Stop-and-go patterns that propagate backward through traffic
Phantom jams: Congestion that appears without any obvious cause like accidents
The point to take from this is that a jam can appear with no crash and no bottleneck to blame, simply because one small decision spreads through the traffic. That gap between a single local action and the pattern it produces is what we mean by emergence.
This example demonstrates the core principles of ABM:
Individual agents (cars) with simple behavioural rules
Local interactions between nearby vehicles
Emergent system behaviour (traffic patterns) arising from individual actions
Spatial dynamics where each car’s position and movement matter
2.1.1.4.2: Boids flocking simulation#
Another classic example of ABM is the Boids flocking model, where individual birds (agents) follow three simple rules:
Separation: Avoid crowding nearby flockmates
Alignment: Steer towards the average heading of neighbours
Cohesion: Steer towards the average position of neighbours
From these simple rules, realistic flocking behaviour emerges:
The Boids model demonstrates the same core principles as the traffic model, now in a different setting. As with the traffic jam, no bird leads the flock and no rule describes the shape it takes; the coordinated flocking pattern is emergent, arising only from each bird reacting to its immediate neighbours. Two very different systems, cars and birds, make the same point: local rules give rise to global patterns.
The simulation below turns that idea into something you can steer: it lets you set the strength of each of the three rules and then release a hawk into the flock. Before you try it, predict whether a coordinated, evasive response can emerge even though no rule describes one.
Because the birds move through smooth, real-valued positions rather than across a grid of cells, this evasive, self-organising flock is also the natural case for continuous space in an agent-based model.
2.1.1.5: Important definitions#
To lock in the vocabulary introduced above, this closing section gathers the key terms in one place. It is a deliberate recap rather than new material: the characteristics from 2.1.1.2 and the purposes from 2.1.1.3 return here as named definitions you can check yourself against.
Important
Key terms:
Agent-based models: Computational simulation models that involve many discrete agents
Bottom-up approach: Starts from the agents and their interactions with the environment
Decentralised, individual-centric: Focus on individual agents rather than centralised control
Adaptive agents: Agents can change behaviour based on experience or environment
Heterogeneous environment/agents: Agents and/or their environment can differ
Purposes: Prediction, explanation, description, theoretical exposition, illustration, analogy, social learning
The key ideas above each carry a name that you will meet again and again, so it is worth checking that the name and the idea have joined up in your mind. An agent-based model is made of many discrete agents, built from the bottom up, running without central control, able to adapt, and free to differ from one another. Fixing this shared vocabulary matters because the rest of the book, and the wider literature, leans on these terms as shorthand. The activity below asks you to match each description to the term it defines, and any mismatch is a signal to reread that idea.
Carry away these five names together, because they are the vocabulary the rest of this book uses whenever it talks about what makes a model agent-based.
With a clear picture of what ABM is, the next section turns to why it matters, and what agent-based models let us do that other approaches cannot.
Key takeaways#
Key takeaways
An agent-based model represents a system as many autonomous agents that follow simple rules and interact within a shared environment.
Its defining characteristics are agent autonomy, heterogeneity, local interaction, and the emergence of system-level patterns from individual behaviour.
ABM serves several purposes, from prediction and explanation to illustration and policy support, and is best suited to systems in which individual detail drives the outcome.