Model-Based Science: A Very Short Guide

A version of this blog post was published on The Conversation on 8th April 2020 and discussed with Amanda Vanstone on Counterpoint (ABC Radio National) on 11 May 2020 (audio available here) and discussed with Simon Lauder (on ABC Radio Sth East NSW) on 10 May 2020).

​
From bushfire prediction to climate change, scientific modeling has been getting a fair bit of press lately, and no more so than with the current COVID-19 crisis. As countries battle the pandemic, scientific modelers are playing a central role in predicting how the virus will spread, what impact it will have, and what sorts of interventions might halt it. 

Whilst the public profile of modeling has perhaps never been greater, the broader understanding of what scientific models are (and what we can and can’t expect of them) remains poor. As a philosopher of science, there is little I can do to contribute to the current effort against COVID-19 apart from stay home. 

One thing I do know a little about, however, is the nature of scientific models, their strengths and limitations. Given that, here I offer a brief and accessible guide to scientific modeling for the uninitiated in the hope that a broader knowledge of their power and pitfalls will be of value in public debate.

What is a scientific model?
Scientific models are representations of parts of the real world. They can take many forms. They range from physical small-scale models of phenomena—such as the San Francisco Bay Model—a hydraulic model used to investigate water flow in San Francisco Bay—to the type of mathematical models used to understand the spread of COVID-19. 

Like any replica or microcosm of the real world, models can be used to indirectly explore the nature of the real world. They can tell us what the important features of real-world systems are, how those features interact, how they are likely to change in the future, and how we can successfully alter those systems.

Why are models so valuable?
Scientific models make it possible for us to explore features of the real world that it is impossible, or impractical, to investigate directly. For example, it would be impractical to do direct experiments on what proportion of the population of Australia needs to engage in social distancing to make “flattening of the curve” likely. 

Even if we could devise good experiments , the time it takes for people to become sick and transmit COVID-19 means any experimental results would arrive too slowly to be of practical use. Scientific models offer us a way to use data from other countries, along with theory and other information to make a reasonable estimate of what impacts particular interventions would have. Models are thus invaluable in a situation like the COVID-19 pandemic where time is of the essence and we are interested in effects at a large-scale.

What are the limits of scientific models?
The usefulness of a model is limited by its accuracy—how well it represents the real world. For example, a model of the spread of COVID-19 based on data from a densely populated part of urban Europe may not work for suburban Sydney or Melbourne because it lacks the relevant similarity to those locations. 

There is a well-known trade off between generality and specificity in model-building. Really detailed modeling such as this work coming out of Imperial College London can be used to make very accurate, relatively narrow range predictions about specific cases in the US and Great Britain. Simpler, general models such as these ones by Ben Phillips at the University of Melbourne, on the other hand, offer valuable large-scale insights, but far less local precision. Such general models have been particularly useful early in the pandemic, when localised information is scarce. However, as we build more information about local circumstances modeling will become more specific and more accurate, and these general models will be less important.

One challenge for modeling in a real-world context like COVID-19 is that our modeling may not get it right every time. This is partly because we lack enough fine-grained information about the real-world situation. It is also because individual actions and sheer bad luck in the short term can make big differences in the longer term. Like a stone thrown into a lake, the failure of one individual to self isolate or quarantine, can produce a much larger scale ripple of downstream effects. The massive impact of individual actions in generating some of the clusters of cases (such as this case in South Korea) is a testament to this.

What does this all mean?
Despite the uncertainty inherent in the COVID-19 pandemic, we should be optimistic about the science. The general principles behind the models we are basing our public policy on are the product of decades of testing and research and we are learning more and more specific information about COVID-19 everyday. In terms of the history of humanity, this scientific progress means we are in a far better place than any generation before us to deal successfully and efficiently with a pandemic of this scale. This is in no small part thanks to the power of model-based science. 


(Thanks to Carl Brusse, Simon Greenhill, Adrian Currie, Ross Pain, Rob Lanfear and Stephen Mann for their comments and suggestions)