h1

Why this could be like 1992 when the polls were simply wrong

April 20th, 2015

Elections & polling expert Keiran Pedley examines whether the UK polling industry could be about to experience a crisis not seen since the polls got it so wrong in 1992.

 Those that follow polling closely will recall the famous situation in 1992 where the polls appeared to point to a Labour victory (of sorts) only for the Conservatives to prevail. Since that time, much work has gone into correcting those mistakes and subsequent election results have shown that the polling industry has been largely successful at achieving this.

However, in an election so close, with so many different parties (and pollsters) involved, could we face the prospect of history repeating itself?  Here I examine some of the challenges pollsters face.

 It’s very close

Aside from the occasional poll showing a 6 point lead for Labour or the Conservatives, most national polls show the two main parties neck and neck. In fact, the current UK Polling Report average has Labour and the Conservatives on 34 points each. Therefore, by understating either the Labour or Conservative vote slightly, the polls could point to a completely different result to the one we see on election day.  Indeed, some pollsters could easily call the election wrong (or right) by sheer accident of ‘margin of error’. What is clear is that the apparent closeness of the race exacerbates the potential for the polls to be ‘wrong’.

 Dealing with a ‘multi-party’ system

In 1992, pollsters were mainly concerned with Labour, the Conservatives and the Liberal Democrats. Now, each of the SNP, UKIP and the Greens are important too. These parties are not just important for their own sake. Polls that overstate UKIP’s support could also understate Conservative support and if Scottish polls overstate support for the SNP then Labour’s situation north of the border could be better than current polling suggests (though still very bad). Therefore, not only is this election close but pollsters are also having to contend with several new variables this time around which are all important.

 Return of the ‘shy Tories’?

One of the major issues in 1992 was the so-called ‘Shy Tory’ phenomenon where voters told pollsters that they intended to vote Labour (or someone else) where in fact they actually intended to vote Conservative. If such a phenomenon repeats itself this could be crucial in such a close race. Former Conservative MP Rob Hayward has produced analysis (more here) showing that polls have regularly understated Conservative support this parliament when compared to the actual result in elections.

Could this happen again? It’s possible. A key group to watch out for is the cohort of Labour supporters that consistently say that they would prefer David Cameron as Prime Minister. For example, the latest Ipsos Mori political monitor shows that some 28% of current Labour supporters say that they are dissatisfied with Ed Miliband’s performance as Labour leader. Miliband’s ratings are of course improving during the campaign and it is debatable how much impact leader ratings have anyway. However, if we are looking for evidence of ‘Shy Tories’ then current Labour supporters unconvinced by Miliband would be an obvious place to start. We should remember, not all party ‘supporters’ in opinion polls are committed activists – some are floating voters liable to change their minds.

 Emergence of ‘shy Labour Scots’?

What about Scotland? A rarely considered factor in this election is whether there are actually a few ‘shy Labour Scots’ north of the border. I admit that this is largely speculation but, such is the frenzy that has gripped Scottish politics since the independence referendum, this is at least possible. This does not mean that Labour are suddenly going to beat the SNP in Scotland but that the scale of the defeat might not be as bad as expected – with significant implications regarding seats won by the SNP.

Another issue in Scotland specifically will be turnout. Scottish voters regularly tell pollsters that they are more likely to turnout than those in the rest of the country. However, an interesting subset of data in the recent TNS Scotland poll here shows that 18% of current SNP voters did not vote last time. If a large proportion of this group stay at home then this could significantly impact how many seats the SNP take from Labour in May too.

How much support does UKIP really have?

Arguably the biggest challenge faced by pollsters is calling UKIP’s vote share correctly. Barring some significant polling convergence between now and voting day someone is going to be very wrong here. UKIP’s support in opinion polls ranges from anywhere between 17% (Survation) and 7% (ICM) at present. Clearly, these both cannot be right. Also, like the SNP, UKIP relies on a significant number of previous non-voters for its support. Again, as with the SNP vote, it will be interesting to see if this support turns up on the day.

Returning to the differences in UKIP support between pollsters, it is interesting to compare levels of support achieved online compared to telephone polls. Below is a recent chart produced by Anthony Wells from YouGov that illustrates this point, consistent this parliament, that UKIP tend to achieve greater levels of support online. Perhaps online polls are correcting a ‘spiral of silence’ among UKIP voters or perhaps they are overstating UKIP support. We will know soon enough. The point is that they are different and both cannot be right.

ukpr

Source: UK Polling Report

Impact of survey mode: telephone vs. online

This brings us neatly onto the potentially critical comparison of online and telephone polling more generally. Recent analysis by Adam Ludlow at ComRes appears to show the Conservatives narrowly leading in telephone polls with Labour narrowly leading in online polls. Time will tell whether this phenomenon is real and continues but it is a potentially critical point. Survey mode is just one of many differences in how pollsters produce their voting intention figures of course but the prospect of one of the main survey modes being ‘discredited’ this election could have substantial ramifications for the UK polling industry in the not too distant future. Personally, I am yet to be convinced but I am certainly watching closely.

Conclusion: A different kind of 1992?

What is clear is that the election is close and pollsters face a number of challenges in correctly reflecting voting intention in national opinion polls. There appear to be some differences when considering survey mode, at least when considering UKIP and the sheer number of parties involved present real challenges. Any one of them being significantly out creates a potential problem. Finally, the age old problem of potential ‘shy Tory’ voters could rear its head again whilst turnout in Scotland could also have a significant impact in the eventual result too.

You could be forgiven for asking whether or not this matters. Many will point out that in this new era of multi-party politics that national vote share only tells us so much about the eventual result. Such is the complicated, often bizarre, relationship between the number of votes cast and seats won. They would be right of course but national polls are still important. They show trends in public opinion and perhaps more importantly all of the forecasts out there right now are based in some way on what the national polls are saying. Bluntly, if the polls are wrong, then everything we think we know about the outcome of the election is wrong too.

For what it is worth I think the polling industry faces a different kind of ‘1992 moment’. Post-election there are likely to be some’ winners’ and ‘losers’ in the polling industry in terms of individual companies. I am also expecting a fair bit of analysis on whether online or telephone polls are any different and which is more reliable and also on how accurately pollsters called UKIP’s support. In the longer term however, the key challenge for pollsters will be to stay relevant. National vote shares only tells us so much and so we should expect more and more regional / constituency level data to be produced in future. This can only help us better understand elections in years to come but also introduces more opportunity for error too. For example, Lord Ashcroft data is regularly used in election forecasting but rarely has competitor data to compare to.

On ‘Shy Tories’, potentially the most important point, we just don’t know.  However, if the industry does experience another ‘Shy Tory’ 1992 moment then expect to hear a lot more from Lord Foulkes in the coming weeks and months and his plans to introduce a statutory body to regular the polling industry. Then things really will get interesting.

Keiran Pedley is an Associate Director at GfK NOP and presenter of the podcast ‘Polling Matters’. He tweets about politics and polling at @keiranpedley.