Statistical modelling

Book review: ‘Escape from Model Land’ by Erica Thompson

Image credit: Frank Gärtner/Dreamstime

How mathematical models can lead us astray, and what we can do about it.

Over the past few years we’ve been inundated with political catchphrases masquerading as unbiased policy which claim to be neutral because they ‘follow the science’. The idea behind the slogan was that – as with all mantras – if repeated regularly enough, people will come to believe it. When the public realised that ‘follow the science’ was only there to elevate baffling and often contradictory politicking to scriptural incontrovertibility, the spin doctors beefed up the linguistic illusion of credibility by adding the word ‘model’. Chanting this, they anticipated, would lead the public to eventually think: ‘well, if there’s a model involved, it must be true.’

As Erica Thompson says in her brilliant ‘Escape from Model Land’ (Basic Books, £20, ISBN 9781529364873), the problem with models is they aren’t all they’re cracked up to be. Readers of E&T will already know that the calibre of the mathematical model is only as good as the quality and completeness of data that goes into it. Further to which, there needs to be an assessment of the legitimacy of its design and interpretation. All of this is covered by Thompson, who is to be congratulated on how well her extended metaphor of the model-as-place holds up: “if we want to ensure adequacy for a certain real-world purpose, we must ensure we enter the right part of Model Land.”

What the general reader may not realise is that models can be as biased as the political media gurus. They can also lead to disastrous outcomes when the makers and interpreters get things wrong. In a world where our understanding of, and reaction to, issues such as global health, climate change and international finance seem to be governed by models, isn’t it time, wonders our author – who is a senior policy fellow at the London School of Economics’ Data Science Institute – that we knew a little bit more about how they work?

The result is a highly engaging work of popular science, in which accessibility to the ideas is given priority over academic muscle-flexing and Thompson explains how our current modelling system is inherently biased towards a particular view of the world. Without understanding how models can be flawed and what their limitations are, we can be led down the garden path into distorted thinking: the precise outcome they set out to avoid. In Model Land, prediction can become fiction, while accountability evaporates, because when outcomes go bad it had nothing to do with politics. It was the scientists and their models.

To understand what models are, how they evolved and why we rely so heavily on them, we need books like this. In tracking their evolution, Thompson says the main reason we persist in using them today “is their undeniable past success.” And yet, she concludes, past success doesn’t relieve us of our responsibility to making the best of our ‘imperfect knowledge.’

Sign up to the E&T News e-mail to get great stories like this delivered to your inbox every day.

Recent articles