Polling Matters is an independent, non-partisan podcast providing expert polling news and analysis, with guests, in the run up to the General Election. In the most recent episode, the team spoke to Chris Hanretty from electionforecast.co.uk and a reader in politics at the University of East Anglia. We discussed his forecasting model and the challenges faced in projecting the most unpredictable election in a generation.
You can listen to this week’s polling matters podcast here.
On this week’s podcast episode we were delighted to speak with Chris Hanretty from Election4castUK. It really is an episode worth listening too (if I do say so myself) as Chris explains how he and his colleagues have put together a model to predict the next General Election. Their current numbers indicate Conservatives 284, Labour 279, SNP 37, lib Dems 27 and the DUP on 8. If that comes to pass in May, expect a long and drawn out negotiation period before a Government is formed. In any case, Chris and the team have created a highly sophisticated model using UK polling data and it will be fascinating to see how close they come to calling the result correctly. You can find out more about their model here.
But here’s a thought. If Chris and the team are forecasting using current polling data – what if the polls are wrong? In this week’s episode, I argue that the polling industry faces the biggest risk of another ‘1992 moment’ since the time itself. 1992 of course being the election that the polling industry got so catastrophically wrong.
The reason is simple, this election looks like it might be very close and there are several new considerations to take into account such as the SNP, UKIP and the Greens. Therefore, if the national polls are out a few points here and there then they might project a very different result than is actually returned.
There are several reasons that this could happen. The first is sample related. Since 1992 much work has gone into ensuring polls conducted are ‘politically representative’, meaning that the polls reflect the country’s political composition rather than merely its demographic makeup. This was perhaps the biggest issue in 1992. Typically, this is achieved by methods such as taking party identification into account or by weighting to past vote recall. However, how might we correctly weight to party identification when the party system is in such a state of flux? Furthermore, we know that the last election is now almost 5 years ago. Just how reliable is past vote recall at this stage? Typically, we know, not very much. If pollsters get the above wrong, it could cause all sorts of problems, especially I would argue when it comes to calling UKIP correctly, a party who will have a great deal of say in the eventual result.
There are other issues too. One is how to ask voting intention questions at all, should we prompt for UKIP for example or treat them as an ‘other’ party. Pollsters disagree on this. Another is methodological. Some pollsters conduct voting intention polls by telephone and some online. In the past telephone polls have been more accurate at calling General Elections but it will be interesting to see if this remains the case this time. Generally speaking online polls have shown UKIP at greater levels than on the telephone, this is not always clear cut but in the latest ICM and IPSOS Mori phone polls UKIP have been in single digits. I am not aware of a time where online pollsters have shown that in the past few years. In any case, some pollsters show UKIP in the high teens, some in single digits, if this is the case in May, someone will be wrong.
Finally, perhaps the biggest challenge of all remains dealing with the so called ‘shy Tories’, those that intend to vote Conservative but will not admit that they will do so to pollsters. Former Conservative MP Rob Hayward has produced analysis here showing that on several occasions this parliament the Conservatives have done better in election results than the polls had suggested. If this happens during the General Election we really could face another polling ‘1992 moment’ where the polls show Labour on course to be the largest party only for the Conservatives to prevail.
So it is clear the polling industry faces several challenges in calling the national vote share right this time around and there is not always consensus on how to achieve this.
Of course, even if they do so, how this relates to the result in terms of actual seats is what is most important to the election outcome. This is why the work Chris and the team at election4castUK is so important. Perhaps then the biggest challenge of all for pollsters beyond the next election will be to stay relevant. They will need to produce far more regional data. We are seeing some of this already with polls in marginal seats, Scotland and the production of England only figures. Expect this to be a permanent fixture of polling in the future as pollsters try and marry that ever diverging conflict between national vote shares and seats in a fracturing party system. Looking at the national figures alone will only tell us so much.
Keiran Pedley is an Associate Director at GfK NOP and presenter of the podcast ‘Polling Matters’. He tweets about politics and polling at @keiranpedley.