Exit Poll 2019

[minionofmidas]

Jan 2, 2020 13:45:21 GMT clyde1998 said:

Dec 14, 2019 14:44:14 GMT minionofmidas said:

Absolutely. Even looking at the seat predictions, there are a number of screwups but most of them are explainable.

Their model evidently couldn't take account of huge personal votes unwinding in Ynys Môn and Norfolk North, both of which were blithely predicted to be 90%+ certain to be holds.

They didn't do too greatly in Scotland, not only overestimating the SNP but also underestimating the Scottish LDs and particularly overestimating the SNP in the North, leading to Edinburgh W, NE Fife, Moray, and Caithness & Sutherland all at 99% or even 100% (Caithness) certain to go SNP and E Aberdeenshire not far behind (and Orkney & Shetland at 81%).

They also overestimated the remain vote in Central London (leading to 97% Lab hold Kensington, and Westminster & City too close to call) and underestimating it in other Southern towns (leading to 88% Con gain Canterbury, as well as numerous "too close" calls that ended up not close.

They mostly modelled the Labour catastrophe across the Northern English ripped out heartland remarkably well - there were no Tory gains here that they were blindsided by even though there were several that they thought less than 50% likely and a few nearmisses they viewed as safe Labour. They did however completely fail to predict the Labour holds in Halifax (Con gain in the 90s, forget the number), Alyn & Deeside (Con gain 97) and Weaver Vale (Con gain 99), making these three arguably the biggest surprises of the night.

IIRC, there are only around 100-200 polling stations surveyed for the exit poll, which will make the Scottish estimate much more susceptible to error than England's. There may only be one polling station in Scotland in a SNP-Lib Dem constituency

Quite so. 144 polling stations of which just 12 were in Scotland (and none in Northern Ireland). Subsample madness is to be expected - same with the London vs provincial southern England thing. That the exit poll was so nearly accurate regarding the Con seat figure (and thus majority over all other parties) was thus quite a bit of dumb luck; some difficult-to-prevent errors cancelling each other out.

[nodealbrexiteer]

Jan 1, 2020 21:09:52 GMT boogieeck said:

Dec 13, 2019 12:06:39 GMT The Bishop said:

In actual fact, the exit poll slightly overstated the Tory victory in both seat and vote terms (the latter was actually 45-33 rather than 46-32)

It has entered legend that the poll was correct. In fact, it was quite a long way off , and in Scotland to an extent that makes a big difference.

The authors of the poll admit cancelling errors in their methodology but I would call it harsh to be described as way off and as with all the exit pols from 1997 I think it revealed all the big messages about the result

[Admin Twaddleford]

All the BBC exit polls between 1992 and 2019:

[nodealbrexiteer]

Jan 6, 2020 23:04:09 GMT Admin Twaddleford said:

All the BBC exit polls between 1992 and 2019:

Thank you-someone else has put one on for October 1974 to 2017 I think. Pity it's pretty much impossible to do one for ITN!

[nodealbrexiteer]

Dec 16, 2019 10:09:24 GMT Robert Waller said:

I don't think the individual seat projections in the exit poll was a very good idea, as it detracts from the impression of its overall accuracy. I note that in the BBC coverage they didn't show Jeremy Vine's 'wall' much as the night went on. I know of at least one viewer who thought that colouring it in meant the seat had actually changed hands, for example Cities of London and Westminster. Sky didn't use the seat predictions, not sure about ITV.
I believe they only polled in 144 constituencies, leaving over 500 with no sample at all.
The idea is not to produce any kind of national vote share, but to look at the situation in different types of marginal contest.
A key element is that they use the same polling stations as last time and calculate changes in these.
Inevitably they will miss factors that affect individual seats.
The only real overall weakness was Scotland. As John Curtice pointed out, they had few polling stations there and were far less confident about the situation 'north of the border'.

I also thought they were slow in updating the overall projection, for example the SNP was still projected to get 52 when it was obvious that the would get fewer.

Overall, though, the methodology still worked pretty well.

Just out of interest when you worked on the ITN/Harris General Election exit polls how would you say there seat projection methodology worked please? A uniform swing projected across the relevant type of marginal seats or was the model more complex like the BBC's?

[Robert Waller]

A uniform swing projected across the relevant type of marginal seats, plus a few adjustments for individual seats that didn't fit general patterns. It worked well in 1987, but in 1992 it slightly underestimated the Conservatives, though the decision was made to say this suggested a hung parliament with C largest party.
With hindsight, I would now recommend we had just said 'C largest party' at 10 pm in 1992, and waited for some actual results to come in...

[nodealbrexiteer]

Apr 29, 2020 20:15:39 GMT Robert Waller said:

A uniform swing projected across the relevant type of marginal seats, plus a few adjustments for individual seats that didn't fit general patterns. It worked well in 1987, but in 1992 it slightly underestimated the Conservatives, though the decision was made to say this suggested a hung parliament with C largest party.
With hindsight, I would now recommend we had just said 'C largest party' at 10 pm in 1992, and waited for some actual results to come in...

Thank you! I always understood ITN used a simpler technique. I would imagine ITN did the same in October 1974,1979 and 1983? And what of the mystery ITN February 1974 forecast I mentioned in the 'Interesting electoral facts' thread(and indeed the BBC one that year)

I guess in 1987 it was a case of tweaking figures to allow for some Tory gains from Labour and the Alliance. I don't know if you should beat yourself up over 1992. The seat forecast was a little closer than the bBC with their more elaborate model and bear in mind the marginals poll like the BBC's found a bigger swing there than nationally. Though Jon Snow says at the start of the programme something like 'we had had thought of not making a seat projection as it's so close' and he did emphasise the better Labour performance in the marginals. 1997 appears to be the only election since nationwide exit polling started where one of the channels didn't broadcast it's seat projection(BBC).

Mind you when the BBC used other sources of information like the poll of polls in 1979 instead of a marginals exit poll I wondered what information they gleaned rom it to arrive at their forecast. Also in October 1974 why their official on screen forecast was not computer based and done by Professor Michael steed but the authors of the forecasting programme(Dr Clive Payne Professor Philip Brown) that Professor Graham Pyatt was using on air did their own computer forecast based on the exit poll and came closer(but this forecast was not officially aired)