Looking towards the final polls
We are rapidly heading towards the final round of polling before the election - I expect most will emerge tomorrow evening, though we could in theory have some final calls tonight (there will be YouGov and ComRes polls tonight, but I expect they will both also have polls tomorrow).
The final prediction polls get a great deal of attention and they are essentially what pollsters are judged upon. After 1997 and 2001 ICM used to take great pride in calling themselves Britain's most accurate pollster based on their final polls for those two elections. Nick Moon of NOP is very proud of getting the 2005 bang on (though NOP no longer do regular voting intention polls in the UK). To some extent, this is all rather unfair. As we all know, polls are subject to a margin of error and the unavoidable random nature of sample variation. In 1997 ICM were closest through merit - they had introduced methodological innovations that other companies hadn't and duly performed much better than their rivals. In 2005 all the pollsters were well within the margin of error and which company happened to be the closest was basically down to sheer luck - it was NOP, but had their respective random samples been slightly different it could have been any other company.
Spare a thought for the company that has excellent methodology, but happens to get a freaky sample for that final important poll - or consider the pollster with ropey methodology, who is lucky enough to see their biases cancelled out by sample error in their final poll! A good example of how much of this is down to luck is Populus in 2005. Their final published poll had figures of CON 32%, LAB 38% and LDEM 21% - quite close to the actual result (33/36/23), but the largest error of any of the main five pollsters. They did a second poll privately and with slightly later fieldwork for Lord Ashcroft, which has final figures of CON 32%, LAB 36%, LDEM 23%. Rightfully enough it was the one in the public domain that Populus were judged upon, but if they had been the other way around Populus would have been regarded as the second most accurate. Such is the cruel hand of chance!
The other thing worth nothing is that because so much rides on these final polls, pollsters give them extra care and attention. There are often extra adjustments - in 2005 YouGov factored in turnout for just their final poll and MORI reallocated don't knows according to newspaper readership. Pollsters sometimes do final callbacks to check if people have changed their finds since fieldwork started. My favourite example from the past is an election when MORI asked how likely people were to vote, then how likely they were to vote if it rained, and factored in the regional weather forecast. These polls often have extra adjustments they don't normally and, it has to be said, they do sometimes tend to converge upon one another.
So, to our runners and riders. Looking at the averages over the last week or so YouGov, ICM and Populus are all showing very similar figures - the Consevatives around 35% and Labour and the Lib Dems very close together around 28%.
YouGov CON 34, LAB 27.8, LDEM 28.6
ICM CON 34.5, LAB 28.5, LDEM 27.5 Populus CON 36, LAB 27, LDEM 28
TNS BMRB and ComRes are broadly in line with these, ComRes's most recent polls have been showing the Conservatives a bit higher and Lib Dems a bit lower - normally in third place. TNS are polling less regularly, but their last poll still had the Lib Dems up at 30.
TNS BMRB CON 34, LAB 27, LDEM 30 ComRes CON 36, LAB 28.8, LDEM 26.5
Then we have the newer online companies. Generally speaking we have seen them produce consistently different figures, with Labour lower and others higher (though in Opinium's last poll they showed a 3 point boost in Labour's score, bringing them more in line with the phone pollsters).
Opinium CON 33.5, LAB 26.5, LDEM 27.5 Harris CON 32.5, LAB 24.5, LDEM 31 Angus Reid CON 34, LAB 23, LDEM 29.5
Finally there is Ipsos MORI, who have only carried out two GB polls during the campaign, with the most recent appearing to be an obvious rogue. I don't know whether MORI will be carrying out a final GB poll at all, and have no idea what it would show if it did.
My own expectation is that the result will be around the figures ICM, Populus and YouGov predict (probably about where they are now, though it is still possible that things could shift further in the last couple of days) - they have the most solid records and, in my opinion, the most robust methodologies. I am very dubious about the higher levels of other support found by the newer online companies, and what seems to be the consequential lower levels of Labour support, and without any track record to back it up and no record of the establised pollsters vastly underestimating support for others, my expectation is that they will be wrong on this front.
As to which pollster actually gets closest to the result - as I said above, a lot of it will be down to plain luck, and what random sample error does to the final samples. A lot will also depend on what adjustments the pollsters make to their final polls, and to what extent do the companies converge together.
