ICM - Remain and Leave level pegging, and dealing with fast respondents
ICM put out their weekly EU online poll today. Topline figures were REMAIN 45%, LEAVE 45%, DON'T KNOW 10% and tabs are here. ICM have tended to produce some of the most leave figures and the neck-and-neck result actually follows on from a series of polls showing a small leave lead, but this is due to a change in methodology.
As is often the case, ICM's poll is actually less interesting than Martin Boon's commentary that accompanies it - Martin's response to the polling errors of last year has been one of the most candid and interesting of the pollsters, if occasionally one of the most pessimistic. In his article today he writes about what he considers to be the bleak future for phone polling given how people use the telephone these days, but also writes about the problems ICM have encountered moving most of their EU referendum polling to online.
Specifically Martin writes about how when ICM set online surveys live on Friday nights they get a rush of fast respondents that are skewed towards Leave. These entirely fill some of the demographic quotas set for the poll, meaning there is no room for the slower responding Remain respondents. To tackle this ICM have made two changes - one is to their sampling (they will spread it across the whole weekend, rather than opening it fully on Friday), the other is to weight respondents by how quickly they respond. According to Martin's commentary the overall effect of this is to improve Remain's relative position by four points.
I should point out (as Martin does in his article) that this is very much an issue to do with the way ICM carry out their online polls. Other online companies won't necessarily do things the same way or face the same issues. I can only speak confidently about YouGov's systems, but I know YouGov's don't invite respondents to specific surveys, they have a system that sends invites to respondents automatically, ensuring a constant flow of respondents into YouGov's system. This means when people click on their invite (be it immediately, after a few hours, or days later) they are sent into whatever YouGov surveys are open at the time and need someone matching their demographics - hence when YouGov surveys go live they get a mixture of both fast respondents and slower respondents, who may have actually been invited before the survey was even written.
There was also a new contribution to the ongoing mystery of the gap between online and telephone polls on the EU referendum, this time from Prof Pat Sturgis (the Chair of the recent BPC/MRS inquiry into the polling error) and Prof Will Jennings (who served on the inquiry). It can be read in full here, but Pat and Will conclude that "While there are of course many caveats required here, this comparison suggests that the true picture may lie somewhere between the two modes, possibly somewhat closer to online. At the very least it suggests a good deal of caution is needed before concluding that one method is right and the other wrong. That will only be known for sure on June 24."
