Systems with human stakeholders are typically difficult to understand or manage. Examples include, economies, land-use systems, or society itself. The difficulty in understanding such systems largely originates from the variety and complexity of human behaviour. Interventions in such systems may have unexpected consequences, because stakeholders may respond in various and unpredicted ways. Models that we use to describe these systems often fail to account sufficiently for such social factors, may thus lead to wrong conclusions. Agent-based modelling (ABM) aims to improve this situation by explicitly modelling stakeholders as agents that interact with each other and with the rest of the system. ABMs allow us to represent the system such that we can experiment with various assumptions on the behaviour of agents. These rules may incorporate existing theories on social behaviour, and include the capacity to adapt to surroundings, or to learn from previous experiences. Due to its flexibility in modelling a wide range of agent behaviours, ABM has grown to be increasingly popular in fields like ecology, economics, and sociology.
To use any kind of simulation model as a tool to learn about the real-life system, we need to analyse its outcomes. This analysis can be especially challenging for ABMs. Since ABMs contain many interacting agents that can learn and adapt, the outcomes of the model are typically complex. Furthermore, ABMs typically generate large amounts of data, both on the level of the agents and of the system as a whole. This increases the challenge of extracting important patterns from this data to help us understand the system. Thus, the utility of ABMs depends on the availability of methods to analyse their outcomes.
The main method for analysing ABMs is sensitivity analysis (SA). SA is aimed at identifying which parameters in a simulation model are the most influential on the behaviour of the model. Conventional SA methods, however, are not well-suited to deal with the complexity of ABMs. For my PhD, I have worked on the development of SA methods that are specifically intended for ABMs. These methods may help us to better understand the behaviour of ABMs, and thus increases their utility for studying real-life systems. For example, I have proposed a method for analysing the effects of adaptation in ABMs. Based on a simple test-case, this method reveals that adaptation is a crucial for systems to deal with pressures or stresses. Without adaptation, such pressures may cause drastic consequences, or even a complete collapse of the system. Adaptation gives the system the flexibility to reorganise in response to pressures, while maintaining essential system characteristics. Thus, adaptation may be a relevant factor to consider how real-life systems, such as ecosystems or fisheries, may respond to pressure.