Escape from model land: how mathematical models can lead us astray and what we can do about it

Thompson, a senior policy fellow at the London School of Economics, debuts with an eye-opening account of the limits and uses of mathematical models. Thompson explains that models are metaphors for the real world, and that it's crucial to avoid taking them too literally. "Force equals mass times acceleration is the 'correct model' to use to solve the question" of when a truck would reach 60 mph, for example, but real-world conditions contain variables that the model can't account for. Thompson offers a host of lessons, among them that every model depends upon value judgments to determine what's included in them, that models should be understood as "not an objective mathematical reality, but a social idea," and that models contain the biases of those who make them, so increased diversity among modelers is essential for "greater insight, improved decision-making capacities and better outcomes." Thompson wraps up with a list of principles for "responsible modelling," including deciding "to what purpose(s)" models should be applied, and if "decisions informed by this model will influence other people or communities" who weren't considered or consulted in the making of the model. The result is a thoughtful, convincing look at how data works. (Dec.)

Math-based models have become the secret machinery of our society, and this book draws back the curtain for a close look. Reality has a way of confounding expectations, especially those of mathematicians concerned with models. Thompson is a respected statistician whose research focuses on the use of computational models to inform decision-making. In this deep exploration, she delves into the ways in which they have, in many cases, taken over our lives. They are everywhere, from finance to social media to sociology to pandemics to weather forecasts. "If data is the new oil," writes the author, "then models are the pipelines--and they are also refineries." However, as Thompson shows, more raw material has not necessarily created models that are better at predicting outcomes. Though data scientists love to keep adding more and more data, this can make a model less robust and more vulnerable to excluded factors. There is also the issue that models inherently reflect the values of those that build them, even if the modelers fail to recognize it. "Mathematical modelling is a hobby pursued most enthusiastically by Western, Educated, Industrialized, Rich, Democratic nations," Thompson writes. "WEIRD for short." Often, problems arise not from a model itself but from the communication of its output. With climate change models, for example, there is a tendency of advocates to seize on the worst-case scenarios to garner media attention. The scientific veneer of the process can easily turn models into weapons to attack opposing views. As a response to these problems, Thompson proposes a series of questions to ask of modelers in order to identify biases, assumptions, and structural weaknesses. In the end, policymakers should be willing to include model predictions in their thinking, but they should leaven them with their own experience and judgment. The author, who clearly understands her field well enough to point out its limitations, offers sound guidance. A complex subject rendered in accessible terms, with good advice for using models without drowning in data. Copyright (c) Kirkus Reviews, used with permission.

Thompson, a senior policy fellow at the London School of Economics, debuts with an eye-opening account of the limits and uses of mathematical models. Thompson explains that models are metaphors for the real world, and that it's crucial to avoid taking them too literally. "Force equals mass times acceleration is the ‘correct model' to use to solve the question" of when a truck would reach 60 mph, for example, but real-world conditions contain variables that the model can't account for. Thompson offers a host of lessons, among them that every model depends upon value judgments to determine what's included in them, that models should be understood as "not an objective mathematical reality, but a social idea," and that models contain the biases of those who make them, so increased diversity among modelers is essential for "greater insight, improved decision-making capacities and better outcomes." Thompson wraps up with a list of principles for "responsible modelling," including deciding "to what purpose(s)" models should be applied, and if "decisions informed by this model will influence other people or communities" who weren't considered or consulted in the making of the model. The result is a thoughtful, convincing look at how data works. (Dec.)