Modelling Epistemic Uncertainty Within Agent-based Models

Epistemic uncertainty is largely ignored in agent-based models due to the difficulty of implementation in complex simulations. Some believe introducing epistemic uncertainty provides limited benefits and argue that the same outcome can be achieved by treating the agentbased model as aleatory inside a Monte Carlo simulation. We show that it is possible to implement epistemic uncertainty inside an agent-based model via intervals (one of the simplest representations of epistemic uncertainty) under certain circumstances. Our analysis shows that, although Monte Carlo can be reasonably used to discover common outcomes within simulations, Monte Carlo is poor at representing the possible extreme outcomes that might arise. In contrast, an approach using intervals can reveal these possible extremes which are sometimes important to planners.