nodealbrexiteer
Forum Regular
non aligned favour no deal brexit!
Posts: 4,447
|
Post by nodealbrexiteer on Oct 19, 2020 18:51:07 GMT
Prof Colin Rallings on exit polls-ITN election podcast:
|
|
nodealbrexiteer
Forum Regular
non aligned favour no deal brexit!
Posts: 4,447
|
Post by nodealbrexiteer on Oct 20, 2020 15:56:40 GMT
It still seems to me that Prof Rallings and his colleague Prof Michael Thrasher on Sky News use simpler techniques to update the forecast compared to the BBC
|
|
|
|
Post by pragmaticidealist on Oct 20, 2020 18:14:04 GMT
Prof Colin Rallings on exit polls-ITN election podcast: I wonder who the unnamed male presenter is who doesn't like to be told the result beforehand. |
|
nodealbrexiteer
Forum Regular
non aligned favour no deal brexit!
Posts: 4,447
|
Post by nodealbrexiteer on Oct 20, 2020 18:56:33 GMT
Prof Colin Rallings on exit polls-ITN election podcast: I wonder who the unnamed male presenter is who doesn't like to be told the result beforehand. Mr Adam Boulton. |
|
nodealbrexiteer
Forum Regular
non aligned favour no deal brexit!
Posts: 4,447
|
Post by nodealbrexiteer on Oct 21, 2020 14:36:11 GMT
I wonder who the unnamed male presenter is who doesn't like to be told the result beforehand. Mr Adam Boulton. The same Boulton who frustrated me on election night 2001 when earlier in evening trailing the programme he said Sky News would have a forecast at 10pm when they had completed their sampling but when it came they gave no details of the polling evidence from around the country it was based on or any seat numbers-just the words Sky News Forecast:Labour Landslide Victory. |
|
nodealbrexiteer
Forum Regular
non aligned favour no deal brexit!
Posts: 4,447
|
Post by nodealbrexiteer on Oct 28, 2020 20:16:57 GMT
playing around with EC I wondered what the 3 1992 exit polls would project using their national vote share figures(bear in mind it projects a fantasy 1992 election of 650 and not 651 seats):
BBC:Con overall maj of 12(331/281/15/23)-based on 3.7% Con lead
ITN:Con overall maj of 28(339/276/12/23)-based on 4.8% Con lead or 24 overall maj on 10pm 4% lead(337/278/12/23)
The S*n:Lab 7 short of maj(294/319/14/23)-based on 3% Lab lead
|
|
nodealbrexiteer
Forum Regular
non aligned favour no deal brexit!
Posts: 4,447
|
Post by nodealbrexiteer on Oct 28, 2020 21:51:35 GMT
same exercise for 1987 based on the BBC/Gallup and ITN/Harris figures:
BBC:Con overall maj of 30(340/269/19/22)-based on 5.0% Con lead
ITN:Con overall maj of 80(365/244/20/21)-based on final 9.8% Con lead or 66 overall maj on 10pm 9% lead (358/252/19/21)
|
|
nodealbrexiteer
Forum Regular
non aligned favour no deal brexit!
Posts: 4,447
|
Post by nodealbrexiteer on Oct 29, 2020 19:21:00 GMT
same exercise for 1979 based on BBC/Gallup 7.7% Con lead and ITN/ORC 6.8% Con lead:
BBC:Con overall maj 65(350/257/4/24)
ITN:Con overall maj 55(345/261/5/24)
AND
2001 BBC/NOP 12% Lab lead and ITN/MORI 13% Lab lead
BBC:Lab overall maj 171 (415/173/42/29)
ITN:Lab overall maj 179 (419/159/52/29)
|
|
nodealbrexiteer
Forum Regular
non aligned favour no deal brexit!
Posts: 4,447
|
Post by nodealbrexiteer on Oct 29, 2020 19:34:02 GMT
October 1974 we have onetrueoverlord below to thank for the calculations: onetrueoverlord.wordpress.com/2018/06/08/clickbait-title-is-this-the-worst-exit-poll-everrrrr/my own take for the BBC figures showing a 10.5% GB lead then 7.9% then 7.8%(without adjusting for Scotland: Lab overall maj 145(390/210/14/21)-10.5% lead Lab overall maj 101(368/232/14/21)-7.9% lead Lab overall maj 97(366/233/14/22)-7.8% lead with Scotland adjustment the first set of figures changes to lab maj of 133(384/204/12/35 other including 21 snp) with Scottish adjustment the second set of figures(based on final Scottish poll figures) changes to Lab maj of 81 (358/227/12/38 including 24 snp) with Scottish adjustment the third set of figures(based on final Scottish poll figures) changes to Lab maj of 75(355/229/12/39 with 24 snp) |
|
nodealbrexiteer
Forum Regular
non aligned favour no deal brexit!
Posts: 4,447
|
Post by nodealbrexiteer on Nov 1, 2020 14:21:28 GMT
I edited an up to date version of the information on the BBC February 1974 constituency exit polls which is on one of the earlier pages of this thread and also see below a revised paragraph for the BBC October 1974(full original post page 97 of Interesting Electoral Facts):
The BBC poll seemed to go wrong due to the polling stations selected not necessarily being representative of the constituency overall and a higher refusal rate amongst Conservative voters(which of course reared its head in future polls too!). The seat forecasts were non computer ones allegedy done by Professor Michael Steed who appeared on the election night programmes and as Hullenedge indicated above and Clive Payne explained to me they were swingometer based uniform swing based forecasts.In fact under a uniform swing Labour would have had an overall majority of 17 and not 3 so the exaggerated Labour lead compounded with the lower majority than expected to multiply the error.
Having corresponded with Prof Steed and him viewing the Oct 74 broadcast yesterday he thinks and I agree that Dr Payne was mistaken in thinking he did the erroneous forecast.
|
|
nodealbrexiteer
Forum Regular
non aligned favour no deal brexit!
Posts: 4,447
|
Post by nodealbrexiteer on Nov 12, 2020 20:20:22 GMT
|
|
nodealbrexiteer
Forum Regular
non aligned favour no deal brexit!
Posts: 4,447
|
Post by nodealbrexiteer on Nov 30, 2020 20:09:49 GMT
ITN/MORI 1997 exit poll seat projections through the day(Projections from MORI Swingo model with 'handset' assumptions in some seats)-(Con/Lab/Lib Dem/Oth-GB only).Notional 1992 result in seats polled 46.6/31.5/20.2.Actual 1997 result in seats polled 34.6/42.1/18.7 so the changes in percentage point terms were -12.0/+10.6/-1.5 or a total vote swing of 11.3% to Lab. The figures next to the seat projections are the percentage changes as forecast by the cumulative interviews to that point then the next number is the number of interviews to that point, the last number is the total vote swing to Lab to that point : 10am:182/408/43/8 (-12.9/+11.1/-0.7)-1991-12.0%. 12pm:183/407/43/8 (-11.6/+10.4/-1.1)-4246-11.0% 3pm:180/411/42/8 (-12.5/+11.0/-1.0)-6445-11.75% 5pm:180/410/43/8 (-12.6/+10.9/-1.0)-7984-11.75% 7pm:170/420/43/8 (-13.1/+11.5/-1.1)-10768-12.3% 8.15pm:178/412/43/8 (-13.0/+11.3/-0.9)-13716-12.15% 9.15pm:180/410/43/8-This was the basis for the broadcast 10pm forecast with slight adjustments by ITN of 177/409/45(with 28 others) (-12.8/+10.9/-0.8)-14888-11.85% 10pm:182/408/43/8 (-12.9/+10.7/-0.6)-15761-11.8% based on vote shares changes as per the article below: www.ipsos.com/ipsos-mori/en-uk/1997-itnmori-general-election-exit-pollThough in the programme Alastair Stewart gave a figure of 12.2% Just bumped this up as I've revised it to show the amount of interviews each prediction was based on, the changes in each parties vote share to that point and the overall total vote (Butler swing) swing to that point. |
|
nodealbrexiteer
Forum Regular
non aligned favour no deal brexit!
Posts: 4,447
|
Post by nodealbrexiteer on Dec 10, 2020 20:09:03 GMT
Who was the fellow poster please?  . Probably missing the bleeding obvious! |
|
|
|
Post by greenchristian on Dec 15, 2020 15:34:42 GMT
Who was the fellow poster please? . Probably missing the bleeding obvious! |
|
nodealbrexiteer
Forum Regular
non aligned favour no deal brexit!
Posts: 4,447
|
Post by nodealbrexiteer on Dec 15, 2020 21:21:04 GMT
Having recently been in email correspondence with Dr Clive Payne who with Professor Philip Brown created the BBC election night forecast computer program that was used broadly unchanged from February 1974 to 2001(though 2001 used in prototype the method used from the joint exit polls from 2005) I can add some comments to my previous posts. Payne refers to forecasts made before any results as 'prior forecasts' reflecting the fact that not all were based on an exit poll. The Payne/Brown program was really to make results based forecasts however they created prior forecasts using whatever polling data there was available or that the BBC had commissioned so that they could counter what he calls 'declaration bias'-for instance early declarers maybe being predominantly Northern safe Labour seats so not forming a representative sample of seats. The prior forecast would therefore help to stabilise the early forecasts and also gave the forecasters something to start with. Unlike the ITN forecasts which were all broadcast from October 1974 onwards and therefore were clearly for public consumption(though his articles suggest ITN made a prior forecast in February 1974 which wasn't broadcast) it's clear to me that this wasn't the case and it seems the BBC really only got serious about highlighting the prior forecasts as entertainment and to compete with ITN from around the 1983 election. Also I got from him that the Payne/Brown forecasts weren't always the ones the BBC used on air-for example the disastrous October 1974 landslide forecast wasn't done by the Payne/Brown team but they did use the exit poll data to produce their own computer based prior forecast which had a much smaller error(28 seats) in the forecast of the winner over the second party. The February 1974 prior which had 6 seat error in the gap between the 2 largest parties clearly wasn't broadcast and to my mind the 1979 prior forecast that the Payne/Brown team did(4 seat error in the forecast gap between the 2 big parties) differed from the on air one which at 10pm forecast Con 315-339,Lab 269-293,Lib 5-13,SNP 1-5,Other 12-15 implying for Con and Lab midpoint forecasts of 327 and 281 which would equate to a 24 seat error in the gap. Payne also asserted the 1970 prior he did(or was involved in using a rather different mean swing based methodology) was different to anything the BBC broadcast-though I wasn't aware of any prior forecasts before February 1974. He also said the BBC used a different prior in 1997-I think by this he meant as I posted before that it wasn't broadcast-the BBC on air used the 13% swing forecast by the national vote shares to imply a forecast of 'nearly 200' based on that swing being uniform. Unfortunately Payne didn't keep a record of the forecast seat totals for the prior forecasts the Payne/Brown program produced in Feb 74, Oct 74 and the possible alternative one in 1979. Also he emphasised to me the 3 'different' types of exit polls as he saw them: 1: Exit polls in individual marginal seats that were possibly representative of similar marginals(BBC did them in 1,3,2,2 seats in 1970,Feb 74,Oct 74,1979 respectively and ITN in 2 seats in Feb 74). 2: Exit polls in marginal seats such as Con/Lab and Con/Lib Dem seats to estimate vote share changes in these seats from the last election and then used to produce estimated seat tallies-basically the ITN method Oct 74-2001. 3:The methodology from 2005(though as I said used by the BBC in prototype form in 2001) where 1 polling station in each seat(mainly marginal but not all) would be polled to estimate changes in share at that polling station compared to the last election and almost treat them as a subset of real results. Statistical models would be then used to identify explanatory variables to help generate a seat prediction-for example it would seem in recent elections things like the size of the Leave vote in a constituency. The Payne/Brown methodology and the one from 2005 onwards basically allows the team to estimate shares for each party and then convert them into the probability of each party winning that seat. The overall seat tallies are the calculated by summing each parties win probability for each seat. So for example let's say in seat 1(ignoring third parties) the Con probability is 0.7 and the Lab one is 0.3 and in seat 2 it was Lab 0.7 and Con 0.3 then in aggregate across these seats you'd expect them to win 1 seat each. Probabilities would also be calculated by information such as expert opinion for seats that were classed as 'special'-eg ones that couldn't be forecast from an exit poll or declared results from similar seat types-examples of these were ones maybe where Plaid had a chance of winning, seats that changed hands in by elections or where rebel independents were standing against the main parties. Also see the 2 articles below(though beware some of the figures in the errors in the BBC and ITN prior forecasts are not quite right from the data I've gathered over time-for example BBC 1997 should be 25 not 29 and BBC 2001 16 not 6.ITN Oct 74 should be 12 not 14,1983 should be 32 not 34,1992 should be 54 not 70,1997 should be 22 not 21 and 2001 should be 16 not 10): pdfslide.net/documents/statistical-methods-for-election-forecasting-in-the-united-kingdom-197090.htmlwww.ea.sinica.edu.tw/eu_file/12015700884.pdf |
|
|
|
Post by Andrew_S on Dec 15, 2020 21:31:48 GMT
The main reason I've compiled running totals spreadsheets for every election from 1974 onwards is because I wanted to see how many results you usually need to get the final changes in share within a certain level of accuracy.
For example if at almost every election you only need about 50 results to get the final changes in share within 1% and 100 results to get it within 0.5%, you can probably assume that'll happen again at the next election.
|
|
nodealbrexiteer
Forum Regular
non aligned favour no deal brexit!
Posts: 4,447
|
Post by nodealbrexiteer on Dec 15, 2020 21:35:40 GMT
The main reason I've compiled running totals spreadsheets for every election from 1974 onwards is because I wanted to see how many results you usually need to get the final changes in share within a certain level of accuracy. For example if at almost every election you only need about 50 results to get the final changes in share within 1% and 100 results to get it within 0.5%, you can probably assume that'll happen again at the next election. Pardon me if I've missed something but where are these spreadsheets please? |
|
|
|
Post by Andrew_S on Dec 15, 2020 21:39:53 GMT
The main reason I've compiled running totals spreadsheets for every election from 1974 onwards is because I wanted to see how many results you usually need to get the final changes in share within a certain level of accuracy. For example if at almost every election you only need about 50 results to get the final changes in share within 1% and 100 results to get it within 0.5%, you can probably assume that'll happen again at the next election. Pardon me if I've missed something but where are these spreadsheets please? I'll post them shortly. You must be fairly new to the site if you haven't seen them before. |
|
nodealbrexiteer
Forum Regular
non aligned favour no deal brexit!
Posts: 4,447
|
Post by nodealbrexiteer on Dec 15, 2020 21:41:32 GMT
Pardon me if I've missed something but where are these spreadsheets please? I'll post them shortly. You must be fairly new to the site if you haven't seen them before. Not that new lol! But there's probably a load of older threads I need to view! |
|
|
|
Post by Andrew_S on Dec 15, 2020 23:07:53 GMT
|
|
. Probably missing the bleeding obvious!
