Some thoughts on what I might say
A week today I’ll be travelling to Westminster where I have been invited to give evidence before the House of Lords Committee that’s carrying out a review of what went wrong with GE2017 polls.
Depending on the questioning by their Lordships I expect to make the point that the campaign had been dominated by the CON landslide narrative that had been reinforced by the May 4th local and mayoral elections where the Tories enjoyed their best night for ten years. That very much supported the general election polling at the time with CON leads of upto 20%
UKIP’s virtual wipe-out in the May 4th elections had created the widespread assumption, wrongly as it turned out, that the Tories would pick up the lion’s share of UKIP’s GE2015 votes and, accordingly, we could assume blue gains in some very unlikely constituencies.
A second theme that ran very strongly was, based on GE2015, that the polls always understated the Tories. So if there was a range of GE2017 polls with similar fieldwork dates then the ones with the best CON position should be given the most credence. The polls that had it much closer like Survation and the YouGov model were dismissed as outliers.
It was in this context that many of the pollsters felt confident that their post GE2015 changes, particularly on turnout modelling, had been correct.
The mood just before the election was very much reflected in the Andrew Neil’s interview with Survation’s Damian Lyon Lowe about his polling that pointed to a hung parliament.
I am hoping to make the point about my long-standing view that leader ratings (not the best PM numbers) should be given much more emphasis and have in the past proved more reliable.
Thus in June the leader ratings had moved sharply away from Mrs. May to Mr. Corbyn in the final few days. This made GE2017 like the GE1992 and GE2015 polling fails when this data had given much better pointers to the final outcomes than the pollsters’ voting intention numbers. Thus at the start of the campaign Mrs. May had a net +17 well/badly score with YouGov. By polling day that was down to a net minus 5. Clearly something was happening.