Today's Britain-wide YouGov poll shows support for the SNP and Plaid returning to 5% for the second time since the start of the year. It also appears that the better showing for UKIP in the last two polls was a bit misleading, with Nigel Farage's party slipping back to 14%. Given the pattern of recent results, it's probably reasonable to conclude that the true position for UKIP is around 15% - assuming YouGov's methodology is basically right, of course.
Britain-wide voting intentions (YouGov, 12th-13th January) :
Labour 33% (n/c)
Conservatives 32% (n/c)
UKIP 14% (-3)
Greens 7% (+1)
Liberal Democrats 7% (+1)
SNP/Plaid Cymru 5% (+1)
The Scottish subsample figures are : SNP 46%, Labour 31%, Conservatives 16%, Liberal Democrats 3%, UKIP 2%, Greens 1%. That's (marginally) the best Labour showing of 2015 so far, but of course what really matters is the gap between the SNP and Labour, and that remains as huge as ever. It gets even worse for Labour, though, because the only reason the gap is as "narrow" as 15 points is the weighting by Westminster-centric party ID, which is particularly extreme in this poll - SNP and Plaid identifiers have been downweighted from 66 to 34. It seems highly likely that without this distorting procedure, the SNP's lead in the subsample would be north of 30%. (And, who knows, the SNP and Plaid might be on 6% or 7% of the Britain-wide vote!)
Aye, running average of SNP respondent down-weighting has hit a new high this morning.
ReplyDeleteCan't help but feel when Ashcroft's constituency polling comes out, we'll see 2010 weighting as standard, even though this is thoroughly discredited in Scotland. Still, would likely make for Labour complacency and hopefully enough detail will be provided for us to estimate real VI.
As pointed out by Patrick Roden on a previous thread and sourced from Hansard.
ReplyDeletehttp://www.publications.parliament.uk/pa/cm201415/cmhansrd/cm150113/debtext/150113-0002.htm
3:52 pm
David Mowat (Warrington South) (Con): Like my hon. Friend the Member for Macclesfield (David Rutley), I will be supporting the charter.
This has been an interesting debate. We have heard principled speeches from the hon. Members for Glasgow North East (Mr Bain) and for Brent North (Barry Gardiner) and the right hon. Member for Morley and Outwood (Ed Balls) explaining the travesty of Government economic policy and saying how bad the economy will get and how the country needs a Labour Government to sort it out. The odd thing is that about half an hour from now those three Members, who spoke in so principled a way about why Government policy is wrong, will go through the Lobby on the same side as me in support of this “gimmick”, this “cheap stunt”, this “travesty”, as it has been called. At least the SNP, the Green party and the Welsh nationalists have taken a principled position. The hon. Member for Dundee East (Stewart Hosie) made an eloquent and reasoned speech about why targets are wrong. We used to say in business: “The great thing about not knowing where you’re going is that you can’t get lost.” That would be a summary of the SNP’s position.
Why is the Labour party going to troop through the Lobby to support the Government? I have only one explanation. I may be wrong, and it is possibly above my pay grade to get involved, but I think that Labour’s decision to support the Government tonight is the start of overtures around a grand coalition. I think Labour has realised that the polls are changing and it is not looking too good out there for it. It has few options left.
Yesterday we saw another example of how we already have an informal Tory-Labour coalition. Seems we are not the only ones to notice the direction of travel.
Ashcroft NHS poll released - massive sample (20,000 across GB, ~2,000 in Scotland), but fieldwork done in mid-November.
ReplyDeletehttp://lordashcroftpolls.com/wp-content/uploads/2015/01/NHS-poll-Full-data-tables.pdf
Westminster VI in Scotland has SNP 46, Labour 24, Tories 14 (page 11).
Correction: after weighing for turnout (page 18), the figures are SNP 46, Labour 25, Tories 15.
DeleteLikely to be false recall issues again as the proportion recalling voting Labour in 2010 is too low and SNP in 2010 is too high.
It could be that some people may be embarrassed to admit to voting Labour in 2010.
DeleteJust taken a new Populus poll with a very bizarre voting option choice.
ReplyDeleteThis was the first screen
http://gyazo.com/421e92a0c92f1852a1eda8d5fc966175
BNP? Has this been done before with BNP as a first choice party?
They're probably just trying out various layouts as testing for polling in the actual General Election and this might not go into a published poll. Some pollsters probably put the BNP on for Euro-election polls - at least in the weeks before - and they might do so again if they stand as widely as they did last time (they won't). Or they might be testing to see if it does make a difference.
DeleteAs someone new to this site and polling in general, could someone explain the reasoning behind the downweighting and what is its intent? It seems like trying to change the results to meet a preconceived result, but I could be wrong. Please explain.
ReplyDeleteThe point of weighting is try to achieve a result that is representative of the place you are polling. You could have an absolutely enormous sample, but if the group consisted unduly of one type or another (men instead of women, Guardian readers instead of Sun readers) then it would not be a fair representation of where you were polling. If that were the case you would most likely end up with a misleading result.
DeletePolitical weighting is one type of this. Most firms use it, but there are some exceptions. What it generally means is that you take the last major election result and try to get your results to tie in with that.
This process is usually fair enough. but it can run into problems if you have what is called "false recall", i.e. a significant number of your sample incorrectly saying who they voted for. This could be a problem in Scotland because you have quite a big chunk of the population who voted Lib Dem or Labour in 2010, but then voted SNP in 2011. If some of those people incorrectly say they voted SNP in 2010, then the SNP rating in a poll would be incorrectly down-weighted because the polling company would think they have too many SNP supporters, rather than identifying that person as someone who has switched.
Pollsters can also have problems if they don't use political weighting. If supporters of one party or another are particularly committed in their support and are therefore more willing to respond to surveys, then support for that party would be over-stated. I think this is likely in the case of the SNP at present. For example, the best result for the SNP in recent polls was in the Ipsos MORI poll (52%), which did not use political weighting.
The polling company has to get a representative sample. If too many SNP supporters are responding then it would make the polls too favourable to the SNP.
ReplyDeleteSo they weigh the numbers back to how the people say they voted in 2010, which is a problem in Scotland as people that voted Labour 2010 & SNP 2011 they might remember that they voted 'SNP' in 2010 and make the poll less favourable to the SNP (false recall)
Populus weights the result by 2010 party ID, which means that the voters that voted Labour in 2010 but now identify with SNP might cause the result to be less favourable to the SNP.
YouGov have the data on what people actually voted in 2010 so should be more accurate but they can't use that data completely so weight by past vote as their panel has presumable gone through some churn. But they use the actual data to check unusual results and for sanity checks.
So, in summary weighting is an art but it is an attempt by the polling companies to get a representative sample.
YouGov weight by party ID in their GB-wide polls. To the best of my knowledge, they only weight by recalled vote/actual vote in their full-scale Scottish polls. The theory with party ID weighting is that a person's voting intention may change over time, but the party they most closely identify with is likely to remain the same, so you can just keep weighting to the same target figures to get a representative sample. That may be the case in normal circumstances, but there has been a sea-change in Scottish attitudes to both Labour and the SNP since 2010.
DeleteI wish to thank those who responded to my question. Your answers were most helpful. As I now understand it, if the polling company is still weighing by party support based on the 2010 election results, then they are finding higher numbers for Labour and LibDems due to their much higher vote in 2010 than currently. If I have this right, thank you. If not, please explain further. It seems from your explanation that 2010 recall weighting marks down the SNP suppport and inflates the Labour and LibDem support. If so, Murphy and Labour are in worse shape than indicated and the LibDems are potential disappearing from the map.
ReplyDelete"It seems highly likely that without this distorting procedure, the SNP's lead in the subsample would be north of 30%."
ReplyDeletePut another way, it seems highly likely that without this entirely legitimate weighting procedure the results of the poll would have produced a completely absurd result.
No - what the scrapping of Westminster-centric party ID weighting would do is produce results more in line with YouGov's own full-scale Scottish polls, which for a very good reason do not use that "entirely legitimate weighting procedure". If you think the recent full-scale Scottish polls have been producing "completely absurd results", then I fear you could be in for a hell of a shock on May 8th.
DeleteYou said a 30% lead, not "more in line with YouGov".
DeleteI certainly didn't say "more in line with YouGov", because that would have made no logical sense whatever - how can a YouGov poll be brought into line with YouGov?
DeleteIf you don't understand the topic at hand, it might be better not to leave fatuous anonymous comments, old chap.
I'll add that I see no evidence for any increase in the number of SNP-ID'ing respondents as a function of Scottish sample size, i.e. when polls have been over subscribed in Scotland by too many respondents per capita compared to other UK areas, it is not due to a disproportionate number of SNP IDers taking part. The SNP rise since the referendum is down to people who previously did vote Labour and Lib in 2010 - and admit to doing so - saying they plan SNP.
ReplyDeleteThere is some evidence for a possible slight increase in SNP-IDers in polls since the methodology change. However, these polls are on average a little lower for the SNP so again, no evidence for SNP voters managing to somehow skew polls.
Also, Comres ask an ID question (which party you primarily see yourself as rather than what you plan to vote) and boy has it changed radically since 2010. Labour dominated pre-2010, then, post 2011, SNP and Labour became level. Since the iref, people saying they identify most closely with Labour has collapsed. SNP has made huge gains at their expense. Also evidence that the overall number of people identifying with a party had increased somewhat and solely to the benefit of the SNP and possibly the Greens.
ReplyDelete