Wednesday, March 2, 2011

A few more thoughts on that YouGov poll

I've just been having a belated look at the details of the controversial YouGov poll from the weekend, and there's no great mystery about the two factors that led to a double-digit SNP lead on the raw data being transformed into a handsome Labour lead in the headline figures. The original sample already included more Labour identifiers than SNP identifiers - but the difference was just two (291 Labour to 289 SNP). That was reweighted to become a whopping 478 to 201 advantage for Labour. And on newspaper readership, another crucial aspect of YouGov's methodology, the weighting of the 94 Record/Mirror readers in the actual sample was almost tripled so that they became 251 'virtual' people - who were disproportionately likely to vote Labour, naturally.

In one sense, these figures reinforce the obvious need for weighting - it seems highly likely that Labour do have a greater advantage over the SNP in terms of party identification than the trivial one suggested by the original sample, and it's clear enough that tabloid readers were underrepresented relative to readers of quality papers. But the real issue is where the target figures for the weightings come from, and whether they have any objective basis. YouGov have form on this - after being subjected to persistent criticism in the run-up to last year's Westminster election, they suddenly conceded that their previous methodology had been failing to take account of changes in party ID in Scotland following the SNP's victory in 2007, and that the Nationalists would be "weighted up" by about 2% in subsequent polls. Encouraging though it was to see their willingness to accept an error, it was hard to escape the impression that both the old and the new weightings had to some extent been plucked out of the air on the basis of what 'looks right'. I'd suggest the next lesson they need to learn is that a one-dimensional party ID model simply isn't appropriate for a country with such an entrenched pattern of diverging party preferences for devolved and Westminster elections. If weighting by party allegiance is considered essential due to the difficulties thrown up by YouGov's panel method, a better bet might be to look at how people actually voted in the previous election for the relevant institution, with adjustments to take account of the 'spiral of silence' problem and false recall.

Regardless of what you think of the massive inbuilt 'head-start' awarded to Labour in YouGov polls for Holyrood, one other observation ought to be relatively uncontroversial - the greater the need to reweight certain portions of the sample, the more unreliable the poll results become, even if the weighting system is well-founded. For example, because barely a third of the target number of Record/Mirror readers were interviewed, there's a greater statistical chance that the voting preferences they reported will be unrepresentative of that section of the population, and of course any such distortion will be magnified when the figures are scaled up. The need for such extreme weighting is an inescapable weakness of YouGov's panel method - and yes, James MacKenzie, that remains a valid point no matter how many times the SNP have commissioned YouGov polls in the past.

My own gut feeling - and I said this at Political Betting even after the Ipsos-Mori poll showing the SNP ahead - is that Labour probably do have some kind of lead at the moment, but that the mammoth advantage suggested by the TNS-BMRB poll at the start of the year seemed highly implausible. The same applies to Labour's lead on the list in this poll. Either way, what we all need to urgently remind ourselves of is that this is a PR election, and that the First Minister will be chosen by a vote of all members of the Scottish Parliament, not by a plurality of one party over another in terms of the popular vote or seats. In many ways the most important figure is the combined support for the SNP and Conservatives - not because those two parties are remotely likely to agree a coalition deal, but because the Conservatives are overwhelmingly unlikely to enter into coalition (or indeed any sort of formal deal) with Labour. Even if we assume this poll is accurate, Labour and the Lib Dems in combination have a lead of just 2% on the constituency vote over the SNP and Tories in combination, and 6% on the list. That may put the state of play in somewhat better perspective.


Munguin said...

I don’t understand the MSMs and other Unionist institutions need to give Labour such a commanding lead in what is so transparently a fix. Will that not just encourage the usually more apathetic Labour voters from not turning out on the premise that it is in the bag?

No surprise I think that a YouGov poll commissioned by the Greens shows a tripling of supports for the Greens. Nor should there be any when that same poll is published in a Labour biased paper like the Herald giving the unionists the best of it. I always said that yougov slant their “results” according to who commissions the poll. And I know, being a yougov respondent myself, how biased their questions are. They know for example that I will always support the SNP and that consequently they will come to me if they want a pro-SNP result. Why would it not be otherwise? Yougov are a company that gets paid to poll people. If they consistently put off their commissioners by not giving them what they want to hear they would go elsewhere. Usually however it is not as transparent as it is in this instance.

I agree with you about that TNS poll. It was a flash in the pan. The real result will be considerably closer. But I still think that Labour will clinch it with a narrow margin. It’s the disposition of the other parties that interests me now. I sincerely hope that Labour plus the Lib Dems will still equal a minority Government.

James Kelly said...

Agreed. Based on what happened in 2007, and at Westminster last year, the chances are the Lib Dems would shy away from a formal coalition unless it could command an outright majority (although I suppose there's also the outside chance of Lab-LD-Green).