Quantifying uncertainty in agent based models

Agent-based models (ABM) can be used to simulate the behaviours and decisions of people over time. They have been used in a diverse range of fields, such as simulating consumer behaviour and infectious disease modelling. However, an ABM can never perfectly simulate a real-world process due to a range of uncertainties. For example, there may be observation uncertainty (multiple measurements of the same observation may yield different results), stochastic uncertainty (each run of an ABM may produce different results), and model uncertainty (a model's inability to perfectly represent the real world even when all other uncertainties are removed). There may also be additional uncertainties specific to the process being simulated. For example, pedestrian's may react and move different to avoid others in a dense crowd compared to in a sparsely populated area. Therefore, there will be uncertainty in an agent's behaviour as a result of uncertainty in the environment. This project investigates the presence and effects of model uncertainties, how to quantify them, and how to account for them when calibrating an ABM. By developing an understanding of the uncertainties associated with an ABM, we can obtain more reliable results from the model.