In a week in which the travails of the Labour Party has dominated the news agenda, relative silence on the EU referendum has resulted in a negligible but upward move for the Leave campaigns, up one to 39%. Remain In drops 2-points (43%), leaving it ahead by a fairly insubstantial margin once Don’t Knows are stripped out (53% vs 47%).
Hat tip to the team at the Polling Observatory for a fascinating piece on EU referendum house effects in the Washington Post which demonstrated how much variation exists between different polling companies on this critical next evaluation of polling methods.
Their work is based on ‘poll-of-polls’ analysis, with the table below showing how far each company’s data is above or below the average. Perhaps the most notable starting point is the extent of variation with, for example, YouGov under-shooting the Remain In average by 8-points but Ipsos-Mori over-shooting it by 9-points. It’s difficult to reconcile such divergence without getting into a tiresome methodological debate along the online vs telephone cleavage, so we won’t, but evidently both can’t be right.
Sitting here in the seat most closely associated with the polling company average, you can be assured, gives us little satisfaction. As the Polling Observatory team note with abandon, ‘poll-of-polls’ are no great guide to accuracy, and let’s remember that ICM has historically found great accuracy whilst situated at an outlier position. On those occasions though, we knew why we were the outlier and we wanted to be the outlier.
Here, not so much. We cannot say that we know why our weekly online tracker sits close to the cross-company mean, nor are we sure of where we’d like to ideally find ICM – in contrast to current General Election vote intention polls where we’re very clear in our own minds what we think current polling numbers should be. Of course, the benefit of having been around for five elections’ worth of polling experience helps on that, but like everyone else, we’re virgins at EU referendum predictions.
My own doubts about polling veracity were on full display before the Scottish Independence referendum, and some of the causes of them are being directly transferred across from it to the EU referendum. So while there is a danger here of becoming a polling Cassandra, I’ll still say that when we link the poor performance of online polls before the AV referendum and to a lesser extent in Scotland, there is a brew being mixed here that we’ve disliked the taste of before.
But ICM will still conduct them without compunction. Indeed, we should probably expect the EU referendum to become a fertile testing ground for new polling techniques under 2020 GE consideration, and of course, we have not even started asking about things like turnout in the referendum yet. There is much yet to think about, and do.