Current methodologies lack the sophistication to capture causal relationships between individuals or the resulting feedback that comes about as individuals interact. Agent-based modelling has the ability to simulate individuals, but currently does not accurately capture casual relationships. This project will connect ongoing work in casual inference modelling to agent-based simulations to robustly capture and simulate causal relationships between individuals.
Explaining the science
Agent-based methods are increasingly being used to simulate individuals and to create insight about the impact of their behaviour within complex urban simulations. However, the accuracy of these simulations is highly reliant on capturing the relationships between individuals over different spatial and temporal scales. Current methodologies lack the sophistication to capture these causal relationships or the resulting feedback that comes about as individuals interact.
An area of growing research interest is that of causal inference modelling (methods devised within computing science and extensively used in epidemiology) that is concerned with statistically quantifying the causal relationship between different variables within a population following an intervention (such as the likely population average change in behaviour following a new policy, e.g. free public transport).
Causal inference models are underpinned by a rigorous mathematical framework, which makes it possible to estimate the magnitude of causation rather than simply highlighting correlation. However, causal inference methods are limited in their ability to represent many complexities that characterise individual interactions, such as feedback loops, and spatial dependencies. These processes can be readily simulated within an agent-based model (ABM). ABMs are often built using crude assumptions to formalise the relationship between individuals. This project will connect ongoing work in casual inference modelling to agent-based simulations
- This project will deliver methods that will create detailed understanding about the behaviours of populations.
- These behaviours will be fed into and tested using an agent-based model applied to a smart city problem.
- Software and guidance for applying these methods will be produced.
Understanding and simulating the causal relationships can be applied to any problem where the behaviour and actions of individuals or a population need to be understood. This could forecasting health interventions, testing policing policies or new smart city initiatives.