This site uses cookies. By continuing to browse the site you are agreeing to our use of cookies. Find out more here

OPINION1 December 2014

2015 a crunch test for pollsters

Opinion

Penny Young, chief executive of NatCen Social Research continues the debate about determining voting intentions and takes on some of the issues raised in a previous blog on the subject.

A blog for research-live.com by ICM’s Martin Boon last week highlighted problems facing the pollsters who overestimated the eventual ‘yes’ vote in Scotland’s independence referendum. Martin also comments that pollsters might be overestimating UKIP’s support; and as Survation’s Patrick Brione has analysed, the polls ahead of the Rochester and Strood by-election overestimated UKIP’s lead over the Conservatives by at least five points, sometimes more.

Boon points the finger of blame squarely at social desirability bias: people not admitting what they know they really feel, and the challenge of sensibly correcting for this unquantified bias in a poll. To bolster his case for the phenomenon, he argues that even gold standard random probability surveys like British Social Attitudes suffer from these problems.

As the producers of this long running survey NatCen Social Research acknowledges that every survey can suffer from this problem. But we’re not sure that Martin’s examples support his case.

Take self-reported racial prejudice. It’s not clear why the fact that 2% of the population – (actually 3% in the latest survey – around 1.9 million people) – self identifying as ‘very prejudiced’ should preclude the possibility that 40,000 race crimes were reported in 2010/11. And, in fact, a further 27% admit to at least some racial prejudice – a figure the size of which has truly startled some commentators. 

In my view, the problems of estimating either the ‘yes’ vote in Scotland or the UKIP vote in by-elections are about much more than the shy-voter effect, and illustrate the headache pollsters have in designing weighting schemes, particularly as the political landscape becomes more complex.

The sources of bias are many. Online panels rely on people who have signed up to complete online surveys – individuals who spend their spare time incentivised to fill out questionnaires online can hardly be truly representative of the population. Some telephone surveys are land line only – an increasing problem in eliciting the views of the young.

Should weighting schemes take into account past voting behaviour or not and if so, how to correct for ‘false memory’? And what about correcting for turnout? This is hard enough at the best of times, but in Scotland 16 and 17 year olds were voting for the first time. And the profile of UKIP supporters is still changing which makes it hard to judge the results of a weighting scheme.

All the main pollsters engage in, and comment on, the methodological challenges of their work and argue about how they solve these dilemmas and are to be applauded for that. But even so, it’s not always easy for the public or commentators to judge the quality of different polls.

Whereas British Social Attitudes publishes its response rate ( 54% in the 2014 report), we have no idea how many people a pollster has attempted to recruit before finding someone prepared to participate in a phone survey, or how many people didn’t click through a link to join a panel. This makes it difficult to judge how hard a weighting scheme is having to work to overcome deficiencies in the sample.

Getting polls wrong is a problem in itself, but there are other potential issues. For example, academics have written about the bandwagon effect, whereby electors become more likely to vote for parties and candidates who are likely to succeed. This begs a question about whether over-representing the UKIP vote might influence voters. 

We’ve already overheard Cameron joking that he wanted to sue the pollsters for overestimating the ‘yes’ vote in the Scottish referendum.

My argument is not that polls are a problem. They are at the heart of our democratic debate, and from a commissioner’s perspective, are speedy and great value for money. My argument is that it is important that polls are not consistently proved to be wrong in important ways, or they are going to come in for some criticism for their impact upon our democracy.

The growing impact of UKIP and the SNP on the Westminster outcome, together with the pressure on Labour’s traditional vote, and the apparent decline of the Lib Dems, particularly given the first past the post system, means that the task for pollsters will be more difficult in 2015 and yet their reputation will be on the line. And that’s before we get to an in-out referendum on Europe.

3 Comments

5 years ago

More interesting discussion of a real challenge for the industry. I would strongly agree that polls themselves are not the problem, although your comment that "it is important that polls are not consistently proved to be wrong in important ways" is a bit short sighted. If there are flaws in polls, they should absolutely be called out and discussed - to not do so, in deference to some idea of "retaining the sanctity of polls" is ridiculous. We as an industry need to understand the structural challenges such polls face, as a first step to finding better solutions. If we bury our heads in the sand, we risk becoming redundant, as clients look to 'big data' and other supposed magic bullets to the problems they can see but we won't face.

Like Report

5 years ago

There are two sources of error in opinion polls - or any other form of market research - error arising from imperfections in the sample and error arising from the data collection approach and responses. As Penny Young says, a random probability sample provides some indication of sample quality and the opportunity to correct deficiencies through weighting. There is no objective and stable way of doing so for on-line panel-based research, which is why despite their adoption throughout much of the research/insight business, on-line panels are very little used for either public affairs or media audience research where the findings influence decisions in a very direct manner. Moreover, whilst in principle data collection error is unaffected by sampling methodology, there is also the real possibility that the extensive media coverage given to an 'outlier' such as the YouGov Referendum poll might have the effect of changing the claimed voting intentions in subsequent polls (however conducted) and even conceivably, actual behaviour at the ballot box. So Penny is correct in saying that it is important that polls are not shown to be consistently wrong.

Like Report

5 years ago

What is fairly certain is that, as May 7 2015 draws nearer there will be some flight from the fringes towards the two major parties when people envisage the effect of their vote on a possible new coalition. What will be interesting will be to see this effect being proved or disproved.

Like Report