some random exit polls/election day polls for other ballots

[Georg Ebner]

Feb 20, 2020 20:29:38 GMT nodealbrexiteer said:

AN EMAIL I RECEIVED FROM PROFESSOR ROGER MORTIMORE ON THE MORI 2001 EXIT POLL(I ASKED HIM IF THE SEAT FORECASTS AT VARIOUS POINTS IN THE DAY WERE AVAILABLE AS IN THE 1997 POLL AND HOW THE MORI SWINGO MODEL WAS USED TO MAKE THE SEAT PROJECTIONS):



I’m afraid the projections throughout the day in the 2001 exit poll were never published so far as I know.

You are quite right that national uniform swing is a fairly poor predictor of the number of seats won these days, and for that reason we have never used it in our exit polls. However, uniform swing across smaller groups of similar constituencies is a much better predictor, and in one form or another is the basis of all seat-projection exit polls in electoral systems that do not use proportional representation (although strictly speaking what we use is not swing but change in the parties’ shares of the votes, since there may be more than two parties whose vote shares are relevant). In the case of our 2001 poll, our polling was entirely restricted to marginal constituencies – much of the reason why national uniform swing fails as a predictor is that the swing can be different in safe seats and marginal seats, and of course it is only the swing in marginal seats that matters since the safe seats are won by the same party regardless. Within the marginal seats, we distinguished between Conservative/Labour marginal and Conservative/LibDem marginals, and we measured the swing in each set separately – by using the poll to estimate the swing in Con-Lab marginals and applying that swing uniformly across all those constituencies we got a pretty accurate projection of the number of those seats that would change hands.

In fact we had the capacity to make the model more complex – for example, if the data had suggested that a particular region was swinging differently from the rest of the country, we could have applied a different swing in those constituencies which might have changed the seat projection. (Professor Colin Rallings, who was ITN’s academic consultant, was in overall charge of making the final projection from our polling data and had to make the decision whether to apply any differential swings or not.)

As you may know, for the last two general elections we have combined with Gfk-NOP to conduct a single exit poll which is used by both BBC and ITV (and Sky in 2010), and the academic team that makes the projections from the data is now led by Professor John Curtice. They use more complicated statistical methods to help them analyse the data as it comes in, and they use a more sophisticated “probabilistic” method rather than simple uniform swing, which in effect allows for small random variations in swing within a broadly uniform pattern. But in fact the essence of the method is still the same – different uniform swings applied to several different groups of marginal seats, and the final decision made during election day which groups to treat separately and which to combine. In 2010, for example, they used different figures for the swing in England from the swing they used in Scotland and Wales.

I hope that helps.

Grouping the early incoming precincts into clusters (cities, towns, young/old aso.) is also conducted during ElectionEvenings in PR-countries in order to calculate the FinalResult.

[nodealbrexiteer]

Feb 20, 2020 22:06:07 GMT AdminSTB said:

Feb 20, 2020 21:33:34 GMT nodealbrexiteer said:

And hopefully 2001 too! in 1997 as I said elsewhere they allegedly started with a Labour majority of 175 according to the Nuffield study splitting the difference between the ITN and BBC polls. The new TV Ark shows a photo after 1 result at 10.55pm saying Sky forecast Labour majority 180+. In 1992 I think they highlighted The Sun/ICM exit poll which may be why the misconception that they sponsored it too occurred. 

2001 was a very dull general election compared to the one four years before it, but IIRC on the BBC's election night show they kept a running seats forecast which was constantly updated with each result. I don't recall that happening on any other occasion to the same extent. I think at one point they had the Conservatives on 154 seats, 11 fewer than in 1997.

Yes you do recall correctly about the running seats forecast which started from around 12.15am-i wish they had repeated that.Yes it did have totals between 150 and 160 as did ITN and Sky News-indeed ITN went as low as 143!

[nodealbrexiteer]

Feb 20, 2020 22:16:25 GMT Georg Ebner said:

Feb 20, 2020 20:29:38 GMT nodealbrexiteer said:

AN EMAIL I RECEIVED FROM PROFESSOR ROGER MORTIMORE ON THE MORI 2001 EXIT POLL(I ASKED HIM IF THE SEAT FORECASTS AT VARIOUS POINTS IN THE DAY WERE AVAILABLE AS IN THE 1997 POLL AND HOW THE MORI SWINGO MODEL WAS USED TO MAKE THE SEAT PROJECTIONS):



I’m afraid the projections throughout the day in the 2001 exit poll were never published so far as I know.

You are quite right that national uniform swing is a fairly poor predictor of the number of seats won these days, and for that reason we have never used it in our exit polls. However, uniform swing across smaller groups of similar constituencies is a much better predictor, and in one form or another is the basis of all seat-projection exit polls in electoral systems that do not use proportional representation (although strictly speaking what we use is not swing but change in the parties’ shares of the votes, since there may be more than two parties whose vote shares are relevant). In the case of our 2001 poll, our polling was entirely restricted to marginal constituencies – much of the reason why national uniform swing fails as a predictor is that the swing can be different in safe seats and marginal seats, and of course it is only the swing in marginal seats that matters since the safe seats are won by the same party regardless. Within the marginal seats, we distinguished between Conservative/Labour marginal and Conservative/LibDem marginals, and we measured the swing in each set separately – by using the poll to estimate the swing in Con-Lab marginals and applying that swing uniformly across all those constituencies we got a pretty accurate projection of the number of those seats that would change hands.

In fact we had the capacity to make the model more complex – for example, if the data had suggested that a particular region was swinging differently from the rest of the country, we could have applied a different swing in those constituencies which might have changed the seat projection. (Professor Colin Rallings, who was ITN’s academic consultant, was in overall charge of making the final projection from our polling data and had to make the decision whether to apply any differential swings or not.)

As you may know, for the last two general elections we have combined with Gfk-NOP to conduct a single exit poll which is used by both BBC and ITV (and Sky in 2010), and the academic team that makes the projections from the data is now led by Professor John Curtice. They use more complicated statistical methods to help them analyse the data as it comes in, and they use a more sophisticated “probabilistic” method rather than simple uniform swing, which in effect allows for small random variations in swing within a broadly uniform pattern. But in fact the essence of the method is still the same – different uniform swings applied to several different groups of marginal seats, and the final decision made during election day which groups to treat separately and which to combine. In 2010, for example, they used different figures for the swing in England from the swing they used in Scotland and Wales.

I hope that helps.

Grouping the early incoming precincts into clusters (cities, towns, young/old aso.) is also conducted during ElectionEvenings in PR-countries in order to calculate the FinalResult.

That's interesting. It made me think about the relatively different techniques the 3 channels use to get their initial projections and then update them-ITN and Sky News using simple techniques than the BBC(see my emails from Professors Thrasher and Mortimore I copied and pasted into this thread). In 2019 Thrasher referred to a regional swing model. see page 18 of the article below about half way down for a description of ITN's method www.nuffield.ox.ac.uk/politics/papers/2002/w8/forecast.pdf




 

[nodealbrexiteer]

1997 BBC/NOP poll on the Scottish devolution referendum 1997-2006 voters sampled on the day before polling and polling day overestimates of the Yes votes by 1.3% and 0.7%

Still looking for the Welsh equivalent

[Davıd Boothroyd]

Feb 29, 2020 13:51:41 GMT nodealbrexiteer said:

1997 BBC/NOP poll on the Scottish devolution referendum 1997-2006 voters sampled on the day before polling and polling day overestimates of the Yes votes by 1.3% and 0.7%

Still looking for the Welsh equivalent

Some details via Peter Kellner speaking from 12 minutes.

[nodealbrexiteer]

Feb 29, 2020 13:58:45 GMT Davıd Boothroyd said:

Feb 29, 2020 13:51:41 GMT nodealbrexiteer said:

1997 BBC/NOP poll on the Scottish devolution referendum 1997-2006 voters sampled on the day before polling and polling day overestimates of the Yes votes by 1.3% and 0.7%

Still looking for the Welsh equivalent

Some details via Peter Kellner speaking from 12 minutes.

Thanks for that though I couldn't see a forecast for the yes/no votes which I believe they did give out

[Davıd Boothroyd]

Feb 29, 2020 20:22:34 GMT nodealbrexiteer said:

Thanks for that though I couldn't see a forecast for the yes/no votes which I believe they did give out

I assume they did ask that question. But because the result was obviously on a knife-edge, the BBC were very, very wary about appearing to predict the result. Until near to the end, they did not stick their necks out. But then, with just three results to go, they did - as seen here:

[nodealbrexiteer]

Feb 29, 2020 21:05:51 GMT Davıd Boothroyd said:

Feb 29, 2020 20:22:34 GMT nodealbrexiteer said:

Thanks for that though I couldn't see a forecast for the yes/no votes which I believe they did give out

I assume they did ask that question. But because the result was obviously on a knife-edge, the BBC were very, very wary about appearing to predict the result. Until near to the end, they did not stick their necks out. But then, with just three results to go, they did - as seen here:

Yes i did read about the constant use of the phrase too close to call. The poll is referred to in the link below without percentages-i'll keep trying to find it.I think it suggested a clear but not overwhelming Yes vote

www.walesonline.co.uk/news/politics/story-night-wales-voted-yes-13617114

[nodealbrexiteer]

Having brought David Butler's study of the 1984 European Elections I now have some more details on the BBC/Marplan poll taken on June 13th-14th and the ITN/Harris exit poll:
Actual GB shares Con 40.8%, Lab 36.5%,All 19.5%
BBC share forecast(sample size 2817) Con 38%, Lab 41.5%,All 17%
ITN share forecast (sample size 3740) Con 40%, Lab 36%, All 21%

Unfortunately it didn't have the seat forecasts which I believe I have somewhere still packed up.

[nodealbrexiteer]

After the BBC pioneered exit polling with the Gravesend poll in the 1970 election it was ITN that picked up the baton with by elections exit polls from the early 1970s to the early 1990s and exit polls at the General Elections from February 1974(albeit only constituency exit polls in that election),the 1975 EEC referendum, the 1979 Scottish and welsh devolution referenda. However they didn't do any polls for the 1979 European Election(which I believe they intended to but cancelled due to the expense perhaps due to it being a ballot heavy year?) or the 1989 one.

The BBC were more inconsistent as we've seen doing 3 by election exit polls in the mid 1970s from MORI and then ones led by Vincent Hanna in the run up to the 1987 election followed by some from NOP post 1987 to develop an exit polling technique. The only national exit polling they did between the October 1974 General Election and the 1992 General Election was the NOP one for the 1989 European Election. In the meantime they used 'on the day polls' at some general and European elections. Post 1997 they seemed to have a mini revival in doing 'on the day' polls from pollsters like NOP and ICM for the referenda to set up the London Mayor/Scottish Parliament/Welsh Assembly plus the ensuing elections for the said bodies. Unfortunately though not for subsequent elections.

[Davıd Boothroyd]

In the 1980s it was quite common for the BBC to commission constituency polls in byelections, so they had something to talk about before the result on the byelection special programme. These polls were typically wildly inaccurate.

[nodealbrexiteer]

The article below covers the BBC by election exit polls 1981 to 1986

www.sciencedirect.com/science/article/pii/0261379486900156

The 1987-1992 parliament polls were done by NOP and were usually within the margin of error though initially consistently were biased in Labour's favour.After a change where refusals were deal with by weighting and they switched to a secret ballot even though persistent pro Labour bias disappeared the Tories were never overestimated though the last by election Ribble Valley was spot on to the nearest whole for the 3 main parties-this comes from Nick Moon's book on polling

[nodealbrexiteer]

ITV on the day poll Welsh Assembly election 2016

www.itv.com/news/wales/2016-05-05/exclusive-itv-on-the-day-poll-breakdown/

[nodealbrexiteer]

Yougov on the day re-contact poll EU referendum 2016 Remain 52%, Leave 48%(4772 sample size). IPSOS MORI equivalent Remain 54%, Leave 46%.(Result Leave 51.9%)
Yougov on the day poll Scottish independence referendum 2014-54% No, 46% Yes(1828 people after they voted on the day plus 800 who had already cast postal vote(Result No 55.3%)

[nodealbrexiteer]

article on MORI 1994 European Election exit poll in London for the Electoral reform society:

books.google.co.uk/books?id=zvT7abT7UZYC&pg=PA1&lpg=PA1&dq=MORI+1994+european+election+exit+poll+electoral+reform+society&source=bl&ots=lejwV0P5i-&sig=ACfU3U3D4MPrzzR_NM421JWzwwi0imYeng&hl=en&sa=X&ved=2ahUKEwjfvdyPtLboAhV6VBUIHVsiBpcQ6AEwAXoECAoQAQ#v=onepage&q=MORI%201994%20european%20election%20exit%20poll%20electoral%20reform%20society&f=false

[nodealbrexiteer]

Was just wondering I remember seeing on here that the 2019 exit poll vote shares were 46% Con 32% Lab-did anyone see any figures published for the other parties?

[nodealbrexiteer]

2 documents where the General Election exit poll results have been collected together-the first one from David Cowling does have some erroneous figures but I haven't got contact details for him:

drive.google.com/file/d/1EO3jblUXk2zgfJ5bQuC6MYG9GOBJ1Pcu/view

The 2nd one s from page 445 of Butler's British Political Facts(I was able to email Professor Roger Mortimore to make some tweaks to the next edition):

books.google.co.uk/books?id=RfpmDwAAQBAJ&pg=PA445&lpg=PA445&dq=butler's+british+political+facts+exit+polls&source=bl&ots=59xxQXJ034&sig=ACfU3U1hvZ7Uumor3rrdUDK0rkb7L4LjcA&hl=en&sa=X&ved=2ahUKEwjltrOV2obpAhWEYsAKHXLzBvIQ6AEwAXoECAoQAQ#v=onepage&q=butler's%20british%20political%20facts%20exit%20polls&f=false

[nodealbrexiteer]

on p324 of the Political communications book for the 1983 election the BBC/Gallup poll figures for those who had already voted plus those who were definite to vote are given as:
Con 43.2%
Lab 28.7%
All 25.6%
Oth 2.5%

[nodealbrexiteer]

BBC February 1974 3 constituency exit polls(initial figures released within minutes of 10pm):

Watford:Lab 39.7%,Con 33.5%, Lib 25.4%, National Front 1.4%(actual 40.47/34.48/23.65/1.4)

Bristol North West:Con 42.4%, Lab 35.4%, Lib 21.8%, Ind 0.4%-updated to Con 40.3%, Lab 37.1%,Lib 21.8% (actual 39.8/38.6/20.9/0.8)

Halifax:Lab 39.5%, Con 36.0%,Lib 24.5%(actual 40.93/35.07/24.01)

Sir David Butler quoted the average swing to Labour across all 3 seats the BBC polled(in the final figures) was 2.2%

[nodealbrexiteer]

The ITN February 1974 exit poll in Preston South showed a 3.9% swing to Labour but I don't have the percentages or the figures for Sutton and Cheam(the other seat ORC polled for ITN).