Driven by data; ridden with liberty.
Perhaps it is a common misconception that ‘shy Tories’ were responsible for the 2015 general polling miss. In The Independent, Shehab Khan proffers a new species of voter: ‘shy Corbynites’:
“Shy voters” have become increasingly influential in elections around the world. In last year’s US presidential election and in the UK Brexit referendum, pollsters blamed shy voters for their failure to predict results correctly. Such voters are believed to have said they would vote for Remain, or for Hillary Clinton, when asked but then, in the privacy of the polling booth, cast their ballot in the opposite direction.
This is a falsehood. Polling companies did not blame shy Trump or shy Leave voters.
In their report into 2016 US election polling, the AAPOR identified three principal causes for state polling misses in the presidential election: real change in voter preferences in the final week; failure among some companies to weigh data on education level; and some Trump voters who participated in pre-election polls did not reveal themselves as Trump voters until the election, and outnumbered late-revealing Clinton voters.
This last cause might indicate the possibility of shy Trump voters being a substantial factor, but a number of other tests do not back this hypothesis. For instance, “interviewer administered polls did not under-estimate Trump’s support more than self-administered IVR and online surveys”, which is inconsistent with a ‘shy voter’ hypothesis.
For the EU referendum, the British Polling Council conducted a short inquiry. It was noted by Professor Curtice that a plurality of BPC member polls in the final campaign showed Leave leads.
Indeed, there was no late swing — a change in intended actions after the polling event, than from when those people were reporting it in a poll — at all. This immediately rules out the possibility of ‘shy Leave’ voters in the final polls, since real changes in voting intention and deliberate misreporting would look the same to a recontact survey. The respondents has said they would do one thing but actually did another: the difference between a real late swing and a ‘shy’ voter is the reasoning in the respondent’s head.
Furthermore, ‘shy Tories’ were unlikely to be a contributory factor to the 2015 general polling miss, as the NCRM-BPC inquiry report states:
We reject deliberate misreporting as a contributory factor in the polling miss on the grounds that it cannot easily be reconciled with the results of the re-contact surveys carried out by the pollsters and with two random surveys undertaken after the election.
Next, Mr Khan asserts:
In last week’s local elections, Labour failed to gain control of any local authority and lost 4 per cent of the vote. The result has been analysed as a precursor to the main event next month. But the local elections excluded areas where Labour has been making huge inroads, such as in London for example.
The “vote” share being cited is the BBC projected national share (calculated by Professor John Curtice and Professor Stephen Fisher), a local election performance measure designed to eliminate the locality being cited two sentences later.
Also, Labour do not appear to “making huge inroads” in London, where a YouGov poll conducted between 24th and 28th March 2017 (for Queen Mary University of London) finds Labour at 37%, with the central estimate being seven percentage points lower than their 2015 general election share.
Mr Khan persists:
Meanwhile, pollsters and pundits had expected a total Labour capitulation in local government and that didn’t happen either.
There was no such poll. In fact, a YouGov poll of Welsh local election voting intentions had Labour marginally ahead of the Conservatives. Of course, local elections are different to general elections, and people vote differently in different elections.
Mr Khan continues:
Yet, interestingly, even without factoring in the shy voters, Corbyn is beginning to close the gap on the until now huge Tory poll lead. A YouGov survey showed the Tory advantage of 23 points a fortnight ago has now fallen to 13 and after the local election results; the odds for a Tory majority, although still the favourite outcome, hit its lowest point since the election was called. The lead is being cut.
This is, at best, highly selective. At the time of writing, the latest YouGov poll (sponsored by the Sunday Times, 4th – 5th May 2017) has a Conservative share of 47% and a Labour share of 28%. This is a lead of 19 percentage points.
The first seven polls (covering 18th to 21st April) since the general election was called had an average Conservative lead of 21 percentage points, and the latest seven (covering 2nd to 8th May, at the time of writing) have an average lead of 18 percentage points.
This is a small “cut”, where sampling variability and differential non-response following the snap election call should be kept in mind. Also, the seven polls conducted before the snap general election — which were conducted more sparsely — had an average Conservative lead of 17 percentage points.
This fall in the polling average could plausibly be a regression towards the mean.
Furthermore, it is better for focus on vote shares, than leads, as leads are more volatile.
Finally, Mr Khan concludes:
But we are not about to witness the total annihilation of the Labour Party that many are predicting. Don’t be surprised if the shy voters pull it through.
There are no serious predictions for the “total annihilation” of Labour, just Labour to be substantially reduced in the Commons, facing a large Conservative majority.
Do not be surprised if ghosts fail to appear.