Written and maintained by CASRAI Editorial Board
Last updated
Short answer
An agent-based model (ABM) is a computer simulation in which autonomous agents, each with its own attributes and behaviours, interact with one another and with an environment. Large-scale patterns are not programmed in; they emerge from those individual interactions. Researchers use ABMs as virtual laboratories where direct experimentation may be infeasible or unethical.
Definition and scope
Weimer, Miller and Hill (2016) describe agents as “self-contained objects within a software model that are capable of autonomously interacting with the environment and with other agents,” and describe ABM as building complex models bottom-up by specifying agent behaviours and the environment. They note the same idea appears under other labels: agent-based simulation, complex adaptive simulation systems and object-oriented simulation.
Citing Macal and North (2014), the same paper defines ABM as “a modeling and computational framework for simulating dynamic processes that involve autonomous agents,” and says every ABM has three parts:
- Agents with attributes and behaviours.
- Agent relationships and interaction methods, meaning who can interact with whom.
- The environment the agents act in.
The defining benefit the authors give is that a modeller can observe emergent behaviour arising from individual behaviour. In biology that typically means entities such as cells, organisms or individuals treated as separate decision-making units, but the definition itself is not tied to any discipline.
How an ABM is built and run in practice
An ABM is specified from the bottom up: define the agents and their rules, define the environment and the interaction structure, then run the simulation and record what happens at the system level. Weimer et al. note that data collection and analysis generally take more time, because individual-level output has to be aggregated.
The same paper names four widely used platforms: NetLogo and Repast (free, open source), and AnyLogic and Simio (commercial). For a worked look at one of them, including how BehaviorSpace repeats runs with different random seeds, see the CASRAI guide to NetLogo and BehaviorSpace.
Related approaches: where ABM ends and discrete-event simulation begins
Weimer et al. are candid that the line between ABM and discrete-event simulation is subtle. Their working proposal is that ABM is simulation in which the entities have greater autonomy in movement and decision making than usual. Treat this as a rule of thumb from one introductory primer, not a formal boundary.
Where it matters for labs: the ODD checklist
When you read, review or write up an ABM, the most useful question is whether the description is complete enough for someone else to reimplement the model. The ODD protocol (Overview, Design concepts, Details) exists for that purpose. Grimm et al. (2020, Journal of Artificial Societies and Social Simulation 23(2) 7) set out seven elements; the table below turns them into a review checklist.
| ODD element (Grimm et al. 2020) | Question to ask of a model description |
|---|---|
| Purpose and patterns | What question is the model for, and what patterns is it meant to reproduce? |
| Entities, state variables and scales | Which agents and environment units exist, what state do they carry, and at what spatial and temporal scales? |
| Process overview and scheduling | What happens in each time step, and in what order? |
| Design concepts | Which general design concepts does the model state it relies on? |
| Initialization | What is the starting state of a run? |
| Input data | Does the model use external input data, and where does it come from? |
| Submodels | Are the detailed rules of each process spelled out? |
Grimm et al. report that ODD documents ABMs, is used by some modellers as a design workflow, and supports replication. The 2020 update adds guidance on structuring complex ODDs and a summary format for journal articles, and revises the document structure to cover model rationale and how fitness for purpose is evaluated. The authors also argue that ODD can in principle be used for other types of simulation model, not only ABMs. They also acknowledge remaining limits: ODD documents can become long, and a description without the code can still be incomplete.
For labs that hold simulation outputs alongside bench data, the CASRAI overview of types of research data is a useful starting point for planning how model code, parameters and output are stored. For the wider field that uses this kind of modelling, see What Is Systems Biology?
Common misunderstandings
- “An ABM is just a statistical model.” The framework is a simulation of dynamic processes driven by autonomous agents, not a fitted equation; system-level results are observed from runs.
- “ABM and discrete-event simulation are completely separate.” Even the primer cited here calls the distinction subtle.
- “A published ABM can be reproduced from the paper’s summary.” ODD exists because summaries are often too thin; Grimm et al. note that incomplete descriptions remain a problem without code.
- “Agents must be people.” Agents are software objects; what they represent is a modelling choice.
FAQ
What is an agent-based model in simple terms?
A simulation built from many small autonomous units (agents) that follow their own rules and interact in a shared environment, so that group-level patterns emerge from their behaviour.
What are the main components of an agent-based model?
Agents with attributes and behaviours, the relationships and interaction methods between agents, and the environment (Macal and North 2014, as cited by Weimer et al. 2016).
What software is used to build ABMs?
Weimer et al. name NetLogo and Repast as free, open-source tools and AnyLogic and Simio as commercial tools.
How should an ABM be documented?
The ODD protocol is the widely used standard; its seven elements are listed in the table above.
Sources
- Weimer, C. W., Miller, J. O. and Hill, R. R. (2016). Agent-Based Modeling: An Introduction and Primer. Proceedings of the 2016 Winter Simulation Conference. informs-sim.org/wsc16papers/009.pdf (read 2026-11-08).
- Grimm, V. et al. (2020). The ODD Protocol for Describing Agent-Based and Other Simulation Models: A Second Update to Improve Clarity, Replication, and Structural Realism. Journal of Artificial Societies and Social Simulation 23(2) 7, 31 March 2020. jasss.org/23/2/7.html (read 2026-11-08).
