It is no longer possible to understand politics without understanding polling. It doesn’t just drive the media narrative around politicians and candidates but, often, the policy agendas they adopt and the way they talk about issues. Yet it remains poorly understood, not just by the wider public, but by analysts and office holders too. In the UK, for example, you don’t have to spend long on social media to find an MP promoting an entirely unrepresentative poll from a newspaper website, or a talk radio host claiming a result they dislike is due to that pollster being in cahoots with some nefarious actor.

In this short, valuable guide, G Elliot Morris gives us a brief history of how polls came to play such an important role in politics, and how they work. Its focus is on the US but the debates play out in a similar way in Britain.

The history is interesting, particularly on the various polling gurus used by presidents, such as Emil Hurja, a mercurial small-town hustler who ended up working for Franklin D Roosevelt and transforming the way in which political parties used data.

But the most useful part of the book focuses on the methodological challenges that make polling difficult, and increasingly so. The biggest problem is that people don’t answer phones any more. In the 1970s or 80s pollsters could achieve a representative sample of the population by calling randomised numbers. But now that’s impossible: only a handful of people will pick up and they won’t be typical members of the public.

As a result, polling has moved increasingly online. This has some advantages – it’s much cheaper to collect large amounts of data and easier to do repeat surveys of the same people to identify trends over time. The downside is that companies can’t randomise their sample as they typically rely on people signing up to online panels. This then increases the importance of modelling the sample against ever more complex lists of variables.

It’s when this modelling goes wrong that we see the kind of polling misses that have increased scepticism about their value, even as they become more central to political life. In the 2015 UK election, pollsters overestimated the number of younger voters who would turn out, failing to spot the impending Conservative majority. In 2016 many US pollsters oversampled voters with degrees, making Trump’s victory seem less likely. In 2020 they fixed that problem but again underestimated Trump’s support – possibly because, after he attacked polls, some of his fans stopped answering them.

The reaction to this has been to employ increasingly opaque and sophisticated methods such as MRP (multilevel regression with poststratification). Even the more thoughtful political analysts struggle to understand how these polls are constructed. One result is that there’s little distinction in the amount of coverage well and poorly designed models get.

Morris is surely right in his conclusion that pollsters and the media that use them need to do a better job at explaining complexity and uncertainty. He’s also right that issue polling, where being a few points out matters much less, is more important than voting intention data. It can be a critical tool in pushing back against vested interests by showing the level of public concern about, say, the climate emergency or its dislike of corporate tax cuts. Most of all Morris is right that, for all its problems, polling remains our best tool for understanding how people think about politics. The alternative is prejudice and guesswork.