h1

Why many pollsters overstated LAB so much at the May Euros and what could be happening with current VI polls

July 17th, 2019

 

GE2017 LAB voters forgetting what they did could be causing distortion

After the 1992 polling debacle when John Major’s Tories won an overall majority even though all but one of the pollsters had LAB ahead a big effort was launched by ICM to find out what had gone wrong and we’ve all heard about “shy Tories” less willing to take part in polls.

The firm’s Nick Sparrow in conjunction with Prof John Curtice came up with what is known as past vote weighting to ensure samples were balanced. Basically respondents were asked how they voted last time and their responses were adjusted so that the sample broadly reflected the previous election.

It worked well and for GE1997 and GE2001  ICM became top pollster. At GE2005 another pollster, NOP, used the same approach and got the result spot on.

By GE2010 most pollsters had adopted mechanisms on the past vote model to ensure balanced samples. The only problem is that you cannot rely on those sampled to remember how they voted.  According to an excellent analysis by Anthony Wells of YouGov is what is happening at the moment with many of those who voted for Corbyn’s LAB at GE17.  He writes:

“How to deal with false recall used to be one of the big methodological debates within British polling. Ipsos MORI still don’t use past vote weighting at all because of their concerns over false recall. In more recent years, recalled vote seemed to be closer to reality, and it has become less of an issue. But with the recent major shifts in party support it may once again become a major concern.

At YouGov we have the advantage of a huge, well-established panel, meaning we have many thousands of people from whom we collected past vote data in 2017, before their memory had chance to play tricks on them. Many other companies do not, and must rely on asking people to recall now how they voted in 2017.

This difference may well explain some of the present variation in Labour support between different companies (I suspect it may not be coincidence that the two companies who avoided significantly overstating Labour support in the recent European elections were Ipsos MORI, who don’t use past vote weighting, and YouGov, who are able to use data collected back in 2017 for past vote weighting).

To illustrate Wells did a test with the same data from the sample but processed differently. One using what those on its panel said they did at GE17 and another on how they now recall their vote. As can be seen there’s a marked impact on the LAB share.

The reason for the variation is that the smaller number of those recalling now that they voted LAB at GE17 means that the responses of those who said they did have to be weighted up in order to fit a past vote weighting model.

Mike Smithson