Labour increased its lead last week. So say BMG, Deltapoll, Focaldata, Lord Ashcroft, Opinium, Savanta and Survation.
Alternatively, Labour’s lead shrank. That’s according to More in Common, Redfield & Wilton, Techne, Verian, We Think, Whitestone Insight and YouGov.
Two other companies, Norstat and JL Partners, detected no change.
There are two ways to respond to all this. The first is to dismiss the polls altogether. They are all over the place. Best to ignore them.
I prefer a different approach. Far from confusing us, the proliferation of polls adds to our understanding. Any individual poll is subject to a margin of error. The more polls we have, the easier it is to identify systematic differences between polling companies and changes over time, and to separate random variations from genuine movements.
The result of this is a paradox. There is real doubt as to Labour’s true lead—but none about the broad movements of party support from week to week.
Start with Labour’s lead. The graphic shows the campaign averages reported by the 14 companies that have been conducting polls at least once a week since Rishi Sunak called the election. By combining at least four polls from each company, we go some way to removing statistical wobbles from the data.
Clear differences in methodology separate the three pollsters reporting averages of 16-18 per cent leads from the five with 24 per cent average leads. JL Partners, Opinium and More in Common start with the raw data and adjust it to estimate what would actually happen if the general election was taking place at the time of the fieldwork. In particular they take account of the large number of “don’t knows”, which are heavily skewed to people who voted Conservative in 2019.
I reckon that makes sense, and that Labour’s true lead is less than 20 per cent. However, we cannot be sure. Maybe those ex-Tory “don’t knows” won’t “return home”. Perhaps they are so furious with austerity, Partygate, Trussonomics, dodgy betting and Sunak’s epic misjudgement over D-Day that most will stay at home or vote Reform. If so, the companies reporting leads consistently north of 20 per cent will be proved right.
However, if we cannot be sure of the size of Labour’s lead, we CAN be sure of whether and, if so, how much the election campaign has affected it. The table below shows the weekly average of the results of the same nine polling companies from before Sunak called the election. (Of the 14 in the chart above, five did not start polling voting intention regularly until after the campaign began.)
What this table gives us is a weekly like-with-like account of the ups and downs of the campaign so far. By averaging the nine polls, we can smooth out sampling fluctuations. That average may be skewed for or against particular parties; the point is that the skew is likely to be more-or-less constant. So we can’t be certain that the latest figures—Labour 41 per cent, Conservative 20 per cent etc—are right, but we can be confident that the gap between them has barely moved. Both parties have lost ground since before the election was called, while Reform and, to a lesser extent, the Liberal Democrats, have gained ground.
So, to resolve the conundrum at the start of this blog—seven polls saying Labour’s lead is up, seven saying down and two no change—it’s really quite simple. As the table above shows, there was little or no change between weeks three and four. There is no mystery about seven polls going each side of the line. It’s precisely what we should expect when polls are subject to sampling error. The underlying truth is that the lead stayed much the same. Having lots of polls enables us to say this with much more confidence than if we had far fewer of them.
Too many polls? Nonsense. The more, the better.