Are the Tories really 15 points ahead? Is Corbyn really the most popular leader?

The release of the latest ComRes online poll for today’s Independent on Sunday and Sunday Mirror certainly got people talking. Its toplines showed a 15 point Conservative lead, the biggest since January 2010, the biggest for the Tories while in government since the Gulf war in 1991, the biggest increase in a Tory government’s lead compared with the prior election result since October 1987 (four months after that year’s election) and a tie with ComRes’s June poll for the lowest Labour vote share while in opposition since September 1983, when Michael Foot was preparing to hand over to Neil Kinnock. All of those stats courtesy of Mark Pack.

Twitter predictably went overdrive, with some delighted, some claiming vindication and others crying foul. The controversy centres around the turnout model used by ComRes, which was introduced following this year’s election. Some reacted to the poll by flagging up the polling failiure in May (which for some people has become a stock response to any poll they don’t like) while others complained about the ComRes adjustment. In fact some people even tried to spin it both ways, which in effect means criticising pollsters for their inaccuracy the election and criticising them for trying to fix it! I imagine being a pollster is like being a referee – get it right, no-one remembers, get it wrong, no-one forgets.

But the Comres turnout filter is there for a reason – there is overwhelming evidence (see this, this, this, this and this) that people overstate their likelihood to vote, which shouldn’t be a surprise when final polls routinely show over 90% of people saying they will vote, compared with an actual turnout of 66%, and ComRes felt this needed to be adjusted for. Because the overclaim of likelihood to vote isn’t equal between parties, the filter tends to increase the Tory lead by four or five points, which explains a lot of the disparity between ComRes and other houses.

Subsequent information (as discussed in my assessment of why the polls were wrong) suggests that the turnout effect was probably smaller than this, and indeed ComRes made it clear that this was an interim solution. Though if the turnout effect is smaller, the effect of some other bias(es) against the Conservatives (that pollsters aren’t yet correcting for) must have been larger.

Some have commented that ComRes is “out of line” with other pollsters. This is factually correct, but the insinuation is that it’s thereby wrong and the others are right. This isn’t necessarily the case – it’s possible that the ComRes numbers are “wrong” and the other pollsters are “right”, but it’s also possible that the ComRes poll is, in fact, the most accurate reflection of public opinon. Until permanent solutions are implememented we’ll continue to get a wide range of numbers, though arguably that’s no bad thing on its own – we just won’t be entirely sure who’s right.

Are the Tories really 15 points ahead of Labour?

But while polls disagree on levels, they concur a lot more on the trends. As I’ve written for the Times, new leaders tend to give their party a bounce, and we’re coming up to the exact point at which the bounce should be strongest. Also, as we head away from the election and into mid term, governments usually drop back and oppositions usually gain. Yet the early signs of a Corbyn bounce have faded and, post-Paris, Labour has been heading in the opposite direction – all five of the post-Paris surveys have shown an increased Tory lead when compared with the previous poll from the same pollster (by on average just under two points).

There’s also an interesting disparity in terms of leader ratings. Corbyn scores -28 with ComRes and -3 (the least negative of any party leader) with Ipsos MORI. Since these questions are asked before the turnout filters are applied, the difference is due to something else. That something else is likely to be the question itself – ComRes asks respondents whether they have a favourable or unfavourable view, while MORI asks whether they are happy with the way that person is doing their job.

Something that struck me is that 25% of Conservative voters said they were satisfied with how Corbyn was doing his job, while 80% said the same about David Cameron. Therefore at least 5% must have said they were happy with both (and probably more, given the don’t knows). It’s not entirely beyond the bounds of possibility that a number of Tories have an unfavourable view of the Labour leader, but are satisfied with the way he is leading his party for reasons other than the question intended. There’s no evidence for that, and the reason may well be something else, but it is one way the numbers could be consistent…





About The Author

Related Posts

  • paul

    Although i understand why the turnout adjustment I think people then say its the public’s opinion. This is not then not true, its a predictions of what a GE result would be, not what a representative group of the public think. This has other knock effects, especially the media agenda, which as far I can can tell (I not social scientist, but a chemist….) has debatable consequences. i.e. minster your party is X behind party Y why are you doing such bad job (when using different filter leads to different headline polls and Q).

    • Jonathan Capehart

      Was about to make a similar point, public opinion and election results are two different things. Differential turn out and these days, more and more, differential voter registration levels mean election results will not fully reflect the opinions of the public as a whole, or a representative sample of it. Published data should be clearly split with 2 sets of numbers with one showing public opinion and the other an election result prediction. This would end much confusion, even the article above mixes up the two!

      • https://www.ncpolitics.uk/ Matt Singh

        Not sure which bit was the confused part. The difference between polls and forecasts is very clear – a poll is a measurement of the voting intention, a forecast is a model of voting behaviour at a future date. What turnout models aim to do is measure more accurately the part of the public that votes – they aren’t predicting anything in the “crystal ball” sense. This is just a way of making sure that voters are measured more accurately. Whether it succeeds is another matter…

        Remember too that the voting public has never been the same thing as the entire adult population – some people don’t vote, and including the voting intention of people that wouldn’t vote doesn’t make sense. However their views *are* usually included in questions besides voting intention.

        • Jonathan Capehart

          Where you said, “but it’s also possible that the ComRes poll is, in fact, the most accurate reflection of public opinon.” – your typo not mine – it implies that Comres’s adjustment is an attempt to get public opinion right rather than the electorate’s opinion, no?

          I haven’t been clear either, sorry, as the prediction I refer to is what an election result would be if it happened at the opinion polling date rather than in the future.

          I think it would be helpful if the two were always clearly separated was really my only point, i.e. population weighted sample = public opinion, adjusted to electoral likelihood = election “prediction” and please be assured that I wasn’t looking to take away from what was a very informative article and I suspect one that is close to nailing the reality.