Saturday, January 23, 2010

On Opinion Polls: General Randomised Sample Head Emptying


Two polls then. These are at quite different points in the range of likelihoods. Which are I think The Great Escape, a minimal Labour hold; The Escape, a hung parliament; The Trial, a minimal Con gain; and The Romp, with Faust the poster boy in complete control. For how long nobody knows. Could we even see the paradox of the better he does, the less good his prospects of maintaining the whip hand?

The ComRes one for The Sunday Mirror shows Cameron already on the ropes basically. 4% worse off than the last of their series. Perhaps the Osborne liability coming home to roost? Perhaps the all-round nastiness of yer Goves and yer Graylings? Of Cameron himself? Perhaps just a case of every poll a maverick poll? But, Hurrah anyway!

The other one is far more interesting in lots of ways, though the arguments below apply to both. The ICM "target seats" poll for Murdoch's Tory-friendly News of the World assumes Labour will lose every seat where 4% swing or under does the business for the Conservatives. And then it looks at the 97 next most vulnerable seats where it's Labour incumbent and Tories insurgent. Swings of 4.1% up to 10% required.

Scary stuff when a 10% swing applies. Someone with a majority of 19.9% (say 11,000 or so) loses by a smidge with a 10% party-to-party swing. Consider Manchester Withington 2005 and you're in the right ballpark.

I'm going to include more or less the full legend from UK Polling Report at this point. Followed by a couple of comments I've made. You can also see this, read other comments and join in HERE:

  • The News of the World has a new ICM poll of marginal seats in tomorrow’s paper. ICM’s sample covered the 97 seats where Labour are in first place and the Conservatives in second place, and where the Conservatives need a swing between 4% and 10%. The implied assumption is that seats with a majority of less than 4% are going to be Conservative gains anyway on the current national polls and not worth looking at – instead these are the seats that span from a hung Parliament to a chunky Conservative majority.
  • The topline voting intention figures in these seats, with changes from the last electon, are CON 40%(+9.2), LAB 37%(-7.4), LDEM 14%(-3.8) – so a swing of 8.3% from Labour to the Conservatives. In contrast the last ICM national poll showed a national swing of 6.5%, so once again we find a slightly larger swing towards the Conservatives in the Con-Lab marginal seats they need to win. This has been pretty consistent in all polls of marginal seats in the last couple of years.
  • What it doesn’t tell us is how well the Conservatives are doing against the Liberal Democrats in their marginal seats. On these particular figures it isn’t critical – if there was a uniform swing amongst Lab vs Con marginal seats these figures would net the Tories around 124 extra seats, and added to the 214 seats they start with on the new boundaries that alone would be enough for David Cameron to win a small overall majority even if the Con vs LD battle was completely static.
  • However, for a healthier majority the Conservatives would also need to make some progress against the third party. There is much less evidence of what is happening in Con vs LD seats, and it is much harder to judge what it means when there is, since personal votes and tactical voting plays a larger part and we don’t know for sure how to factor that in.
  • Going back to the ICM poll, there was scant evidence that the larger Conservative swing in marginal seats came from out-campaigning Labour or having more money to throw about. 28% of respondents recalled having received party literature, been canvassed or seen other signs of campaigning by the Conservatives, not significantly more than the 24% who recalled seeing signs of Labour campaigningin their local area.
  • ICM also asked whether people trust Cameron or Brown more on various issues. Cameron led decisively on modernising the NHS, cutting crime, controlling immigration and improving standards in schools and united his party. Brown led on dealing with an emergency. On transport, Afghanistan, taxation, the recession and terrorism the two leaders were almost even.


  • Reading this, with some inkling of what was likely, I asked:

    Where are the numbers polled, level of DK-WS-NV and the methodology for this one? Lack of this hampers evaluation of the importance or not of this. Also any work on squeezability of LDs in these? The old would you rather C or L for this crew could be illuminating. They have 14%. The margin between the real contestants is 3%. Plus it the usual +/- lots margin for error?

    Give us the footnotes, or expect us to shrug and say "who knows what if anything that means then"?


    UK Polling Reports' Anthony Wells responded promptly:

    Chris, they’ll all be on ICM’s website in due course. Basically though –

  • Methodology is all usual ICM methodology. Quoted margins of error figures are always based on the sample size – so with a 1000 sample size this is the usual +- 3% margin of error.
    WNV 10%
    Ref 7%
    D/K 23%
  • Though as usual, ICM reallocated their don’t knows & refusals based on their vote at the last election, so those don’t knows and refusals were not ignored – ICM assume half of them vote for the parties they did last time. The unadjusted figures in this poll were CON 42%, LAB 36%, LDEM 13%. The re-allocation of don’t knows changed the final published figures to CON 40%, LAB 37%, LDEM 14%.
  • 39% of Lib Dem voters said they would consider voting tactically (though no doubt some of those saying Conservative and Labour are already Lib Dem supporters voting tactically), but they were not asked which way they would do so. The closest proxy from the questions asked is probably best PM, where Lib Dem voters split for Brown over Cameron by 43% to 36%

  • Which led me to this response. Earning part of my living administering polls I shouldn't perhaps be quite so cynical and questioning:

    So, basically we have 600 punters who actually express an opinion in 97 seats meaning an average of 6 and a bit per target constituency? Plus some farming out of the DK-WS-NVs.

    There will be serious to very serious over-reporting of voting at all last time as 80-90% is not where we were at turn out wise, plus confusion and mis-remembering of the ones who did actually vote.

    3% margin of error and 5-6% of possibly tactical LDs as well? In some seats I think a very large proportion – more than 39% – will actually go tactical. Seems to me these things are a very blunt instrument indeed.

    Doing quite a bit of phone iD for one of the parties, and face-to-face for one of the majors, not ICM, it beats me why anyone gets too excited by these things. Within the margin of error we have way way lower Tory gains cf ComRes, and we also have landslide territory.

    However this has been carried out there’s nowt as queer as folk. Happened to bump into a guy I polled 2 months ago 1 week ago in another context. He had told me v firmly in the interview that if there were a GE tomorrow he would vote for one of the big two. But when I saw him again, outwith an interview context, he was equally firm that he would in fact vote for the third party.

    And it wasn’t a change of heart. Just a change in candour.


    Polls are an interesting sport of course. Precise. To at least one place of decimals. Yet with a whopping 6% band width. 3% either way. Making a lead from pollster X of 15% OVERLAP in a sense with one from pollster Y saying 10%. Yet one of the leads appears to be fully 50% more than the other! And most journos and politicians scarcely numerate to boot. Never mind the broad population.

    But the combination of this apparent precision (with inaccuracy in the small print) with 24/7 mainstream media and the 24/7 clamour bloggers and tweeters, some with very grand ideas of their own importance, the danger is we get a rather meaningless mash of mush. For the poor woman or man in the street the chance for good old-fashioned sheepism is muchly reduced.

    If so LOL say GOOD. The electorate have some stark choices to make. It would surely be in every one's best interest if every single elector ignored every opinion poll, ignored every gurning televised focus grouper, ignored every bogacious bar chart and speculative swingometer and squealing headline, made no attempt whatsoever to bet on the national winner, and instead voted in the best way available to them.

    To return either a Conservative or a Labour government. In their own interests or in solidarity with those needing the most support from society. A little clue here. This is not those with net estates in the clear blue water zone between some £700,000 and £2 million. And it's not feral hedge fundies either. Or Mum-Ps with no main home and dynasties of unsuitable relations in top jobs.

    2 comments:

    benchilltory said...

    You clearly must be worried by what will happen in about 100 days to spend so much time an effort on this opinion poll.

    In 2005 the poll of marginal seats was very close to what happened in the election.
    One of the opinion polls well worth watching

    http://greattit.blogspot.com/ said...

    Still looks like it could be quite close.