2.1.3: ABM vs other approaches

ABM is not the only way to simulate a system, so how does it compare to the main alternatives, and when is it actually the right choice? This section makes two moves: first it sets ABM beside the other two established simulation paradigms to see where each is strong, and then it turns that comparison into judgement about when the extra effort of modelling individuals earns its cost.

2.1.3.1: ABM vs. system dynamics (SD) vs. discrete event simulation (DES)

Agent-based modelling is one of three main ways to simulate a system, alongside system dynamics, which tracks whole-system quantities as they flow and accumulate, and discrete-event simulation, which advances from one scheduled event to the next. System dynamics describes the world as stocks and flows at the aggregate level (total population, total inventory), so it captures global feedback loops but not the individuals inside those totals. Discrete-event simulation follows entities through a sequence of timed events (an arrival, a service, a departure), which makes it ideal for queues and processes but leaves little room for agents that learn or adapt between events. They differ in the level at which they describe the world, in how much variety between individuals they can hold, and in the kinds of feedback they capture. Knowing where each one is strong matters because choosing the wrong approach can make a problem needlessly hard, or hide the very thing you set out to study. In the table below, click any feature row to read when that feature tips the balance towards ABM, and look for the recurring theme that ABM pays off when the detail of single individuals is what drives the outcome.

The three main ways of simulating a system differ feature by feature. Click a row to see when that feature points you towards ABM.

Feature	ABM	System dynamics	Discrete-event

The point to remember is that ABM, system dynamics, and discrete-event simulation are not rivals but different tools, and the right choice depends on whether individual differences, adaptation, and local feedback matter for the question you are asking.

2.1.3.2: When to choose ABM

ABM is particularly suitable when:

Individual heterogeneity matters

Different agents have varying characteristics, behaviours, or decision-making processes that significantly affect system outcomes.

Local interactions dominate

Agents interact primarily with neighbours, specific groups, or network connections rather than through global mechanisms.

Emergent behaviour is expected

You want to discover how unexpected macro-level patterns arise from simple micro-level interactions without pre-programming the outcomes.

Spatial structure is important

Physical location, proximity, geographic constraints, or network topology directly influence agent interactions and system dynamics.

Policy testing is needed

You want to safely experiment with interventions, compare scenarios, and understand unintended consequences before real-world implementation.

These five conditions describe when ABM fits, but recognising them in a real problem is a matter of judgement rather than a checklist. Reaching for ABM by habit is a common mistake, and so is dismissing it when it would help. The skill worth building is judgement: knowing when individual differences and local interactions truly drive an outcome, so that the added effort of modelling individuals buys you something a simpler aggregate model could not. In the activity below, each case describes a situation. Decide whether ABM fits or a simpler model would do, and let the streak of correct answers show you how consistent your judgement is becoming.

Learning Activity: ABM or simpler?

decide when ABM is worth it and when a simpler approach wins.

Eight short cases shown one at a time, each a modelling situation to be judged as suiting an agent-based model or a simpler aggregate one.

For each case, press either ABM fits or Simpler fits, then press Next to advance to the next case.

Case 1 of 8Score 0   Streak 0

ABM fitsSimpler fits

Next

The lesson is that the question to ask is never whether ABM is fashionable, but whether the differences between individuals and their local interactions actually matter for what you want to learn.

2.1.3.3: Checklist: when is ABM suitable?

Having judged individual cases, it helps to turn that judgement into a repeatable check you can run against your own project. The activity below expands the five criteria from 2.1.3.2 into concrete prompts, applies the scoring guide, and recommends the modelling level that best fits the evidence.

Learning Activity: Is ABM right for your project?

weigh a real modelling problem against the five criteria for ABM.

A twenty-point checklist grouped into five criteria, system characteristics, interaction patterns, heterogeneity and adaptation, spatial and temporal aspects, and research goals, that gauges how well your own project suits agent-based modelling.

Tick every prompt that is true of your own project, then press Show my recommendation. The activity will apply the scoring guide and suggest the most suitable modelling level.

System characteristics

There are multiple individual entities to simulate (people, animals, organisations, etc.) These entities have different behaviours, characteristics, or decision-making rules The system exhibits complex, non-linear dynamics You expect unexpected patterns to emerge from individual interactions

Interaction patterns

Entities interact with each other directly (not just through global variables) Local interactions are more important than global ones Interactions depend on proximity, networks, or specific relationships The structure of interactions affects system outcomes

Heterogeneity and adaptation

Individual differences matter for the overall system behaviour Entities can learn, adapt, or change their behaviour over time Different groups or types of entities exist in the system Variation between individuals affects aggregate outcomes

Spatial and temporal aspects

Location, geography, or spatial structure is important Entities move through space or change locations Neighbourhood effects or clustering phenomena occur The timing and sequence of interactions matter

Research goals

You want to understand how macro patterns arise from micro behaviours You need to test interventions or policy changes You want to explore "what-if" scenarios safely Traditional aggregate models seem insufficient for your problem

How the recommendation is calculated

15–20ABM strongly recommendedIndividual detail and interaction are central to the problem.

10–14ABM likely suitableThe case for agents is substantial, but check data and implementation cost.

5–9Compare approachesABM may help, but a simpler aggregate or event-based model may answer the question.

0–4Start simplerThe checklist gives little reason to represent individual agents.

Show my recommendation

Learning Activity: When is ABM the right choice?

judge when agent-based modelling is the right choice.

Five scenario-based questions that apply the suitability criteria to unfamiliar problems, checking whether your judgement transfers beyond your own project.

For each question, choose the option you think is correct and press Show answer. Read the explanation, compare it with your reasoning, and use Try again if needed. Your selections are saved automatically.

Across all of these checks the deciding question is the same: choose the approach that matches whether individual detail actually drives the outcome, rather than defaulting to the paradigm you know best. You will return to this decision with a fuller framework in section 2.3.3 (Decision framework); for now, with a sense of when ABM is the right tool, the next part opens up what an ABM is actually made of, beginning with its agents and their environment.

Key takeaways

Key takeaways

• Agent-based modelling, system dynamics, and discrete-event simulation describe a system at different levels of detail, and the appropriate choice depends on the question being asked.

• ABM is preferable when individual heterogeneity, local interactions, spatial or network structure, or emergence are central to the problem.

• When only aggregate averages are required, a simpler model is usually the better choice, so ABM should be selected by judgement rather than by default.