Sunday Polls
No sooner have I dismissed October’s ICM poll that showed a 10 point Conservative lead as an obvious outlier (and therefore interpreted their Guardian poll this month as a return to the norm), suddenly another poll appears with a lead of the same sort of size. An ICM poll in tomorrow’s News of the World puts voting intention at CON 39%, LAB 30%, LDEM 20%.
UPDATE: The poll also included an unusual best Prime Minister question. The results were Brown 29%, Cameron 25%, Campbell 5%. Unusually though John Reid was alos included in the question, and was chosen by 8% of people. I also missed last night quite how cretinous the News of the World’s reporting was. The headline was “Nightmare for Cam” – yep, the second largest lead for the Conservatives in an ICM poll since 1992 is “a nightmare” 
Filed under: ICM, Voting Intention
And no sooner had you headlined “Conservatives weakening” . . . and this comes out. There does seem to be some volatility at the moment, but to be having a consistent lead in the polls is not something I’d complain about.
some inconsistency may be down to the spinning that labour are dogged by.possibly,then a poll shows blair leading cameron for the first time in a year.spin may not be dead.this is very unhelpful to the ‘boring brown’ media image.cameron will win with his media genius if the public believe or want to believe him.
A question that has puzzled me.
Are these surveys done UK- wide? Given the extreme difference between English voting and Scottish voting, wouldn’t they automatically show the Conservatives 2.2% points lower than they are [assuming similar voting rates and the national population difference of 10 to 1].
Any comments. Thanks
So what excuses, and more sensible mathematical reasoning, do we give to this massive fluctuation?
Last month there was a massive gap, between different polls, which almost boiled down to the degree of certainty those asked gave to their intention to vote.
The “LibDems” fluctuated between near obscurity and challenging Labour for second place.
Now one poll shows Labour in front, the other with a large Conservative lead, and then another saying Cameron’s position is weakening.
Perhaps there is a wider malaise. Maybe voters are increasingly unclear as to what function political parties play? Maybe they are just being seen as competing PR organisations to which prospective MPs should, or do, pay dues.
Jerry – the overwhelming majority of polls are GB wide, i.e. they cover England, Scotland and Wales but not Northern Ireland. They should show the Conservatives as they actually are in Great Britain, but it is a fair assumpton that they would be slightly higher in England alone, given that they are significantly lower in Scotland.
John – the difference between teh Communicate polls and the ICM and YouGov ones is very straightforward. Communicate do not weight by past vote, and therefore their polls will tend to be more volatile and will tend to be more favourable to Labour.
There is no simple explanation for the differing figures from ICM and YouGov…quite simply the variation is probably due to sample error between the polls, and until more polls come out and the picture stabilises a bit we can’t really tell what the true picture is.
I wonder if Prescott saying at PMQs that Cameron’s honeymoon was over caused the rise.
Perhaps if pollsters had to disclose on all polls how far out they were on the last local or general election we would be able to judge better and they wouldn’t release polls that are plainly outliers.
This Poll wouls suggest that the “others’ have dropped back to 11%, which might suggest that at least part of the difference may be down to how the UKIP and BNP support are feeling about Cameron.
Peter.
I think people are trying to read reasons for changes in poll ratings that are nothing more than sampling variations . I still maintain that there has been no significant change in public opinion since June/July apart from a blip in a rise in Labour support at Conference time . Most other changes in poll figures are due to differing methodology between different pollsters and margins of error . Whether the true level of support for Others is 10/11% as some pollsters show or 15/16 as other pollsters now only time will tell but the variation is clearly tied in with the level of LibDem support and independent of Con and Lab support .
Peter, as Mark said the variance in Others vote seems far more inverse to the variance in Lib Dem vote, not the Tory vote. Low “other” vote, high Lib Dem vote – that seems business as usual. If next months poll has a higher Other vote again then it will far more likely come with a lower Lib Dem vote. I can’t think of anything Cameron has done in months to either appeal or repel BNP/UKIP voters, not since he dropped the pledge to withdraw from the EPP.
Well the Communicate looks like a complete rogue. It pulls down the WMA and makes the ICM deviation 3.9% which is a little harsh, but ICM’s Std is 2.6 so 3.9 is not much outside the range (and without the rogue poll the deviation would be about 2.8). The WMA is boringly 37:32:18 v little change since May.
Peter – “others” haven’t dropped. The high figure was in YouGov’s last poll, and they tend to produce higher figures for others. In the last two ICM polls “others” were at 9%, so if anything they are up.
My personal suspicion is that Mark is right – the only change in party support over the last few months that I am confident of is that there was a short-term boost in Labour support after the party conference which now seems to have gone away.
Ralph – down that way madness lies
When a poll comes out it is impossible to say for sure if it is an outlier. A poll suddenly shows a 5 point drop in one party’s support then it might be an outlier, or the party might really have seen its support drop since the last poll. The problem is, if a poll comes out and shows no difference, then it might be accurate, or it might be that party support has actually dropped (or soared) and it is an outlier that co-incidentally matches the previous month’s poll. Because we don’t know what the “real” score is we can never tell. Once subsequent polls come out we can sometimes make a fair guess, but other than that…
For practical purposes it doesn’t work anyway. It isn’t up the pollster – they can hardly say to a client, “here the poll you paid for but we can’t let you publish it”, or indeed “here is poll you paid for but we don’t like the results so we’re doing it again”. Equally, where do you draw the line between a newspaper not publishing a poll because they think it is an outlier, and not publishing a poll because they don’t like what it shows? I think if the Guardian suddenly decided not to publish next month’s ICM poll because it showed a 15 point Tory lead a lot of people would think they were hiding results they didn’t like. Better that they publish the results, and leave to people to draw their own conclusions.
The detailed info from the ICM poll is now on the ICM site . There is one glaring anomaly/statistical blip in the poll compared to other polls including ICM . The gender split has a fairly expected Con/Lab split of 36/31 in females but for males the split is 43/29 which looks a total rogue .
The website gives a slightly different published voting intention figure to that we have been discussing . It gives Con 39 Lab 31 LibDem 20 Others 10 .
You sometimes get strange figures in the splits. Part of it is the specifics of the weighting regime. LEt me give you an example –
Look at these two weighting regimes.
Regime A: 52% female, 48% male. 50% ABC1, 50% C2DE.
Regime B: 26% female ABC1, 26% female C2DE, 24% male ABC1, 24% male C2DE.
The first one weights by both gender and class. The second one weights by class within gender. If a poll is weighted along the first lines, while the overall total of the poll should be correct, and will indeed have the right gender and class balance in total, the internal splits on those measures might be odd. In theory, most of the ABC1s could be men and most of the C2DEs could be women.
I have no idea if ICM weight by class within gender or by class and gender separately. It’s just an example. YouGov for example weight by age within gender and class separately, so their gender splits won’t be unrepresentative by age – it won’t be all old men and young women or vice-versa, but aren’t necessarily representative of class breaks within gender.
It doesn’t matter for the point I’m making: polls are weighted so they are representative *as a whole*, it doesn’t necessarily follow that their crossbreaks are also representative. You can get odd results on some of the breaks without it necessarily meaning there is anything wrong with the poll as a whole.
Yes I follow that Anthony and often notice some odd figures particularly amongst the smaller subsets but male/female split is pretty basic and the 2 subsets in themselves quite large so I would not have expected such a big variation from the norm especially on something on which there has been a great deal of comment that Cameron is appealing more to women than men .