There will be lots of polling between now and the election, perhaps more than ever before and polling around policy choices will be particularly important when the various parties’ manifestos are launched. It is always worth paying close attention to how questions are asked and looking at a range of polls from different places, not just one, when forming a view on what the public think.
IPSOS, February 20 2024
This election may prove remarkable in a number of ways, not least in that Labour’s lead is sufficiently enormous for it unlikely to be proved wrong on the night. The same was true, of course, for the party’s poll showing in 1997 and 2001, as it was for the Conservatives in 1983.
But the number of abject polling failures suggests this is a deeply flawed practice, nor is there any sign that its accuracy is improving as time goes on, whatever its proponents say. Since 1960, there have been 16 general elections. The polls were broadly correct in 1964, 1966, October 1974, 1979, 1983, 1997, 2001, 2005 and 2010. Nine, in all.
They were wrong in 1970, February 1974, 1987, 1992, 2015 and 2019, six in all. A failure rate of more than 37 per cent.
I’m not sure in what other line of work that would be acceptable to the public, even weather forecasting. But there are compensations for the rest of us, not least in the many and various excuses proffered by the industry for the occasions on which it has fallen down on the job.
One oldie used to be the telephone excuse, in the days before universal access to the ‘dog and bone’. This was wheeled out after Labour’s surprise victory in February 1974. Telephone polling, you see, was biased towards more affluent, phone-owning, thus Tory and Liberal households, thus failed to predict the return of Harold Wilson.
This doesn’t quite explain why the polling business four years earlier had predicted an easy win for the same Labour leader and his telephone-deprived supporters, only for Edward Heath to stage a shock Tory victory.
In parallel, there has been the not-at-home excuse, This is greatly more flexible, in that most people, regardless of income or status, are away from home some of the time. All that is needed is to group into sub-sets – travelling salespeople, working mothers, students and so on – and calculate back from these categories to ‘explain’ what went wrong as a result of their absence from the pollsters’ attentions.
If all else fails, there is always the rewrite-history option, and claim that the polls were spot on. The set-piece example here is the 1987 election, almost never mentioned these days but at the time looking very dicey indeed from the Conservative point of view.
There was great excitement at the thought of a hung Parliament, with the ‘two Davids’ – Owen and Steel – emerging as power brokers. A particular treat was hearing top pundit Brian Walden explain the mechanism by which a government could be put together. A grand functionary called a ‘rapporteur’ would hold intensive talks with No 10, Parliament and Buckingham Palace. Walden shared with his former fellow West Midlands Labour MP Roy Jenkins an interesting take on the pronunciation of words such as this.
Both Margaret Thatcher and her sidekick Lord Young were firmly in the ‘we’re doomed’ camp, with only the party chairman Norman Tebbit insisting all was well. He was right – the Tories returned with a 100-plus seat majority.
Readers may object that I have left out the apparently-burgeoning phenomenon of ‘shy Conservatives’ who either lie to pollsters or refuse to answer. Possibly, but is not the larger problem that people in general prefer not to divulge private information to complete strangers, especially as said strangers are being paid for this exercise and they are not?
Many years ago, my wife and I were watching a very old episode of The Avengers. Not only were Steed and Mrs Peel carrying no identification documents, but they never said for whom they worked or what was their authority, yet people still answered their questions.
It was, my wife suggested, a more trusting era back then.
Not any more it isn’t.
Lion & Unicorn 