Exit Polls

[Maroochydore]

The exit polls published after close of voting predicted a Tory majority of 80. This proved to be spot on.

Polls during the run up to voting day showed a Tory majority shrinking as the days went on (and it was never as high as 80) until it became 'too close to call'.

Does anyone know why the exit poll was so exact and the other polls so widely out? Is it the methodology or something else?

The obvious answer would be; people lied during the run up. Why wouldn't they continue lying during the exit poll?

Any ideas but let's stay away from turning the question into a political discussion.

[PinkDalek]

The exit polls published after close of voting predicted a Tory majority of 80. This proved to be spot on.

I thought the prediction was 86 but near enough. Others might explain more about predictions, polls, probability etc.

Does anyone know why the exit poll was so exact and the other polls so widely out? Is it the methodology or something else?

The obvious answer would be; people lied during the run up. Why wouldn't they continue lying during the exit poll?

People change their minds on the day. Those who get involved in canvassing and "knocking up" will know this.

Any ideas but let's stay away from turning the question into a political discussion.

I'm sure you can read plenty on this at Polite Discussions, should you be willing to access that place.

[SalvorHardin]

A key difference is that people are not asked how they voted.

Instead they are given a dummy ballot paper which they complete. So it is a secret ballot.

People are much less likely to lie on a secret ballot than when having to give an answer to a pollster. Also you are only polling people who actually voted whereas a conventional poll will include those who do not vote.

Voters are picked at random throughout the day at the selected polling stations.

[johnhemming]

It is also after they have voted. People change their minds throughout the process, but once they have voted they remember it for a while. Reliably for a day. The exit poll is within 10 mins.

[mc2fool]

There's an explanation at https://www.bbc.co.uk/news/election-2019-50716626.

It covers the points others have mentioned and has some additional details. One thing to add to it is whereas it says:

"The exit poll is based on 144 constituencies ... The same constituencies are surveyed from one election to the next, for consistency. "

While that's strictly true, it's actually 144 polling stations it's held at, and the reason the same ones are used each time is to provide a baseline (as the UK doesn't publish individual polling station results) for comparison, and what is actually looked at is the change between the exit poll results from the previous exit poll at each polling station, and then that's all fed into some clever statistical analysis to project the national results.

(BTW, this was the reason there was no exit poll for any of the referendums; there was no previous baseline to compare to.)

If you want more nitty gritty there's an article by John Curtice & chums at https://rss.onlinelibrary.wiley.com/doi ... 17.01054.x (only 4 pages, no maths!)

As an aside, I've met John Curtice a few times (always at public lectures, I'm not in the business at all). One time was just before the 2017 election and I asked him if he was doing exit polls again, and he gave a rather nervous sounding "ah, yes", to which, surprised at his tone, I said, "but you're so good at it!", and his reply was, well yes, that's the problem, people's expectations of the exit polls had become pretty high. I watched on the night and noticed that he was pretty equivocal in presenting the exit poll ... got it pretty good though (4 seats off).

[bungeejumper]

In my last small village, polling days would invariably see the wife of our local Mercedes dealer hanging around outside the .... errr, no, serving as a pavement welcomer, or whatever they call it. With her big blue rosette and her clipboard, she would stand there on the pavement ticking off all the party faithful as they arrived, and later on in the day you could be sure that the big fancy cars from the dealership would be going out to ferry the missing members into the polling booths.

And as people came out from the polling station she would shout out, "Good morning bungee, I hope you voted Conservative?" (She was perfectly aware that I hadn't.) I don't suppose it was actually illegal, but I wasn't ever completely comfortable with the feel of it. In a small community, the sense of being watched can become quite powerful. No wonder some people lie.

BJ

[Nocton]

In fact the polls beforehand always showed a good Tory majority and were never 'too close to call' - that would only have been at the extreme end of the probability range. The average difference between Labour and Conservative was always close to 10%, at least as shown in The Times. And The Times map of predicted results about a week before showed a majority of about 50. I think there were two factors suggesting a hung parliament: 1. the slightly pinky/London bias of the BBC coupled with the need to keep the story interesting; 2. clever spin from the Tory electoral machine trying to ensure that prospective Tory voters did not get complacent and not turn out. I live in Lincolnshire, which is now 100% blue. When I saw the Time's prediction map which showed Lincoln remaining Labour - only Grimsby and Scunthorpe turning blue - I was somewhat surprised as I had Lincoln to go back to Tory as almost certain, partly owing to the failure of the local MP to back any form of Brexit. In the end the majority was at the extreme end of the average polls, but still within the probability spread.

[quelquod]

In a small community, the sense of being watched can become quite powerful. No wonder some people lie.

BJ

I know what you mean. In my village even a hint of not supporting the SNP is liable to lead to a nasty scratch on your car at the least. No one dares put an alternative poster in their window! I doubt if a poll would be very accurate.

[gryffron]

The exit poll was:
1) Much larger than conventional polls. IIRC it asked 22k people, the "typical figure for polls is 1k-2k.
2) Regionally fixed. Conventional polls have little idea where the pollees are located. Making it much harder for them to convert vote share to seats. Of course this is particularly important with a FPTP system. Where your voters are really matters. 51% of votes in 51% of constituencies = overall majority.
3) Repeat data. By repeating the exercise at the same polling station, they can, over time, get a much better feel for how the pollees reflect the whole constituency. For example, suppose they are at an inner city polling station in a largely middle-class seat. At this particular station, a vote of 70:30 in favour of Labour might equate to a Tory win in the seat as a whole. But an 80:20 vote for Labour might indicate a more general swing across the whole seat and thus it falling to Labour.
4) Secret ballot, so less likely to lie, or simply not answer, than when asked openly.
5) After the event so no undecideds.

Hope that helps, a bit
Gryff