Strength in Numbers How polls work and why we need them
A lucid defence of the importance of political polling in a modern democracy, G Elliott Morris’s book gives a valuable insight into the process of predicting elections
The result of the Tory leadership election will have surprised nobody – though the catastrophe that followed will have stunned many. The polls all showed
Liz Truss out in front, though by margins varying from four per cent to more than 30. In fact her victory was between these extremes, which meant that virtually every pollster got the margin wrong.
Few honest pollsters will be surprised by this range. For the Conservative leadership was a very different exercise than traditional polls of the general public on (say) voting intention. It is one thing, though not easy, to select 1,000 representatives of the public for a general election poll and then weight the sample for relevant variables such as age, education and sex. It is another to do it when Tory party membership amounts to only around half of one per cent of voters and where it is impossible to know what weights to employ to make sure your sample is representative.
Unfortunately, inaccuracy does not apply only to polls of party members. In the last two American presidential elections, the polls got it wrong. In the first case, 2016, they got the wrong winner – Donald Trump, not the polls’ sweetheart Hillary Clinton. In 2020 they escaped a similar fiasco when Joe Biden only just sneaked in despite polls typically giving him a national lead of around eight per cent. A subsequent gathering of the good and the great of American polling admitted that they had no clear idea why they were so wrong.
This book by The Economist’s American-based data editor G Elliot Morris – how long until every serious newspaper has a data editor? – admits to polling failures. Yet he retains his faith in polling as an aid to democracy. I wonder what he would have made of two recent British polls, one of which showed a large majority of the British public supportive of the railway workers, and the other a near-even split. Which poll should politicians believe as their guide to the people’s will?
Morris’s book is most valuable in its account of the many adjustments the polls have made to their techniques to try to do better. As American polling is vastly larger in scale and funding than British polling, the range is big and intriguing, though the magic spell has yet to be discovered.
For the real trouble with polls is we expect too much of them. There is sampling error – broadly if a poll says Democrat 38 per cent and Republican 38 per cent, within the margin of error this might mean the Democrats are six per cent up or the Republicans six per cent up – or the two
“For the real trouble with polls is we expect too much of them”
are indeed level pegging.
But more recently an even bigger problem has reared its ugly head. More and more people refuse to respond to pollsters – up to 20 “noes” for every participant. So samples require ever-more ingenious adjustments to weight them to the population. Perhaps the miracle is not that the polls sometimes get it wrong. It is that they sometimes get it right.
Given that the author writes for The Economist, this book’s lucidity is no surprise. For everyone interested in polls – and what reader of The House magazine isn’t? – Strength in Numbers is a must-read.