Glasgow Local Group meeting: Agent based models

On the 30th of April 2019, the RSS Glasgow local group hosted a late afternoon event, where Dr. Eric Silverman from the Social and Public Health Sciences Unit, University of Glasgow, talked about agent-based models (ABMs) and their application to public health. The event was attended by around 25 people. Eric's talk was divided into three main sections: an introduction to ABMs, including the different areas of research where it can be applied; an exposition of the application he developed for social care in the UK; and the potential next steps for research in the field. The talk was followed with some interesting questions and discussion from the audience, and a short reception with opportunities to network afterwards.

The introduction to ABMs included a definition of an agent: units of analysis with a certain degree of agency, based on decision rules which allow them to interact with the other units, and the world around them. ABMs are not necessarily complicated, but they generate complex interactions between the agents and their environment. This observed complexity can lead to insights about the underlying processes in a certain scenario, and how these micro-processes lead to macro-level patterns and structures. This is often referred to as 'emergence'.

In this regard, the usefulness of ABMs is in how they are complementary to statistics, and not a substitute for statistical modelling. Dr. Silverman explained that one of the driving ideas behind ABMs is "less computation, more cognition” -- they should be less about number crunching, and more about theory building, since the main tool for analysis in ABMs is the mechanisms through which the agents make their decisions. This is particularly useful when we want to learn something about the individuals and how macro-level properties and structures emerge from the individual dynamics. If we are only interested in the macro dynamics, there is less to be gained from using ABMs.

Some of the different applications for ABMs include:

  • Political geography: Looking at the effects of having a group of refugees settle in a new area
  • Evolution: How the interaction of different selection pressures and environmental/social interactions can drive evolutionary change
  • Synthetic biology: How do microscopic organisms behave and interact with their environment
  • Medical research: Understanding the spread of cancer cells
  • Urban development: To develop traffic simulations, particularly designed to answer questions on traffic jams
  • Entertainment: Procedural generation of cinematic sequences requiring lots of individuals (stampedes, war scenes, etc.)
  • Anthropology: Study of the Anasazi people in South-Western United States, and why they became extinct
  • Political science: voting patterns, including gerrymandering and local elections

The main application introduced by Dr. Silverman is that of the social care aspect of the wider health and social care system in UK. The application has an emphasis on who is providing care to those in need, which is relevant in an aging population. In this particular case, we are interested in three things: How carers balance work, life and care; what is the impact of informal care on overall patient care costs for society; and what are the effects of potential interventions.

The Linked Lives aging model (documentation and actual model available at the project Github page: https://github.com/UmbertoGostoli/ABM-for-social-care), is innovative in including both the supply and demand of care in the population. The model was designed for the UK, where the parameters are calibrated using real-life data. One of the main conclusions from the study is identifying categories of carers that shoulder a disproportionate amount of the care burden, such as the so-called ‘invisible army’ of aged carers helping their spouses.

The ABM approach allows researchers to consider counterfactuals, morally cost-free policy research, and more importantly, how might a particular intervention affect the rest of the system down the complexity chain. In contrast, ABMs are not as useful for predictions: in the same way that understanding plate tectonics does not imply we can predict earthquakes, understanding social dynamics in human populations does not imply we can predict the precise future of society.

The rest of the talk was focused on moving forward from ABMs through meta models to understand consequences over many different scenarios. This is done in conjunction with model emulators, which reduce the computational burden of complex simulations and thus enable more sophisticated analyses of their behaviour. Some questions from the audience followed: what are the differences between meta models?, and how are some better than others?, and how do ABMs allow us to understand the complexities of social care?.

Load more