Making sense of the Tory lead
by Peter Kellner in Commentary, Editor's picks and Politics
Mon January 23, 12:22 p.m. GMT
YouGov's latest poll puts the Tories ahead by 5 points. YouGov President Peter Kellner explains why we can be confident that the Tories really are ahead
YouGov’s latest poll, showing the Conservatives five points ahead, made the front page of the Sunday Times and livened up the blogosphere. No wonder. It is the biggest Tory lead since September 2010, and the first since then to show the Conservatives doing well enough to secure an overall majority in the House of Commons when our figures are translated into parliamentary seats.
And yet, as both I and my YouGov colleague, Anthony Wells, frequently make clear, individual voting intention polls should be regarded with care. A sudden lurch in the figures might be real, or a sampling fluke, or a mixture of the two.
Most of the time, sampling fluctuations matter little. If a poll finds that 75% of the public consider Ethelred to be unready, few people will care if the true figure is 72% or 78% – we can be sure that a large majority wants him to be better prepared. A three-point margin of error seldom makes any practical difference to the conclusions to be drawn from our polls.
Voting intentions are different. The figures that tend to receive the widest publicity are those where small variations have real political significance. So, when our poll for the Sunday Times put the Tories on 41% and Labour on 36%, then – assuming a three point margin of error – the true figures could be Labour 39%, Conservative 38%.
However, I don’t think that is the case. Here are two reasons why.
Firstly, when we say a polling figure has a 3% margin of error, we are operating a statistical convention. Theory tells us that, assuming the poll is properly conducted, the likeliest truth is at or very close to the figures we report. A 2% error is less likely than a 1% error, a 3% error is less likely than a 2% error, and so on.
The good news, then, is that a 3% error is unlikely. The bad news is that 4% error, or even a 5% error, is not impossible. When pollsters talk about a 3% error, they mean that we can be sure 19 times out of 20 that the true figure is within three points of what a poll reports. But there is a one in twenty chance of a rogue poll, which is out by more than three points. So, the chance that the true state of the parties at the time of our Sunday Times poll is that they were level-pegging, or that Labour was ahead, is pretty small. (Remember that the one-in-twenty calculation covers errors in both directions. In other words there is a one-in-forty chance that we significantly overstated Tory support – and another one-in-forty chance that we have badly UNDERstated their true figure.)
The second reason why I am sure the Tories have moved into the lead this month is that we can look at more than one survey. In nine surveys we have conducted within the past fortnight, we have detected Tory leads on five occasions, Labour leads twice, and level-pegging also twice. This is a far cry from our pre-Christmas polls when Labour leads were routine.
Moreover, our last four polls have shown this trend:
- Jan 16-17: Con 39, Lab 40
- Jan 17-18: Con 40, Lab 39
- Jan 18-19: Con 41, Lab 38
- Jan 19-20: Con 41, Lab 36
This pattern indicates one of the advantages of daily polling. If we see a sudden change, especially where there is no obvious reason for it, we are normally able to tell within a day or two whether it is an outlier. This time, our five-point lead fits a pattern. So, although I would not bet the farm on the precise level of the Tory lead, I am now certain, for the first time since December 2010, that the Conservatives really are ahead. Previous Tory leads were narrower, and fleeting, and could be explained by sampling fluctuations. Were we, like other companies, to measure voting intentions only once or twice a month, I would be less certain that we had avoided the curse of the rogue poll.
Why have the Tories taken the lead? Here’s my guess. It has little to do with anything ministers have done. The problem lies with Labour. The stream of bad publicity for Ed Miliband over the past fortnight has damaged the party (as well as his own rating, which continues to deteriorate). Some public sector workers may feel badly about Ed Balls’s tough language on public sector pay. And last week’s spat between Miliband and the unions on this issue may have done further damage. Voters don’t like the spectacle of divided parties. I believe Miliband could retrieve his, and his party’s, ratings if he fights this battle to a conclusion and is seen to win decisively. Then voters would admire his courage and start to see in him a real leader.
Remember Neil Kinnock and Militant: while Labour’s leader at the time was engaged in trench warfare with the far Left, he was unpopular; but when he launched his ferocious assault in Bournemouth in Labour’s 1985 conference, and put his opponents to flight, his ratings soared. In short, voters don’t like internecine battles, but they love decisive victories. Labour and Miliband are suffering from the former, and have not (yet?) achieved the latter.
One final thought on sampling variability. Yesterday morning, radio and television reports spoke of Newt Gingrich’s 'unexpected' victory in South Carolina’s Republican primary election. Nobody watching the polls sensibly would have described it in this way. The pollster.com website listed twenty polls conducted within the final ten days of the campaign. See the website here.
These showed a clear trend, with Mitt Romney’s support sliding, and Gingrich gaining ground. It looks as if Gingrich moved ahead on Tuesday or Wednesday of last week, and continued to surge until Saturday’s vote.
Now, within that clear overall pattern, there were wide disparities between individual polls. Surveys conducted in the first half of last week showed Romney’s support varying between 26 and 37%, and Gingrich varying between 24 and 33%. It would be – indeed, was – simply easy to say that the polls were all over the place. And, given the small samples of likely Republican voters in some of the polls, and hence the increased risk of sampling error, wide variations between individual polls were always likely. Yet Gingrich led in each of the seven polls conducted between last Wednesday and Friday, and by the widest margin in the poll with the latest fieldwork.
Six of the previous seven polls, conducted over the previous three days, had shown Romney ahead, and by an average of nine points. It was perfectly possible to find fault with individual surveys and to treat some of their sample sizes with caution – and yet be confident by Saturday morning that Gingrich had taken the lead and was heading for victory.
Moral of the story: polls should be regarded neither as individually perfect nor collectively worthless. The knack is to distinguish between what can be said with some certainty, and what cannot. And the more polls by the more companies, the better.