|
|
Post by marksenior on Jul 6, 2017 9:32:13 GMT
Analysing the pollstera performance is not as simple as your table shows .
Polling is a several stage process . A sample is taken , it is then weighted by demographic data to eliminate sampling errors and finally pollsters do additional weighting using in house assumptions on turnout and other factors .
If you look at the final polls of the companies it gives a different story on accuracy . Examples
Comres after initial weighting had Con 40 Lab 39 LD 9 so not that far out LDs too high , Con and Lab too low but gap pretty much spot on .
They then introduced their own in house weighting and ended up with a massively erroneous forecast
Ipsos Mori after initial weighting had Con 41 Lab 41 LD 7 pretty much spot on but after in house weightings they produced Con 44 Lab 36 LD 7
ICM after initial weightings had Con 44 Lab 38 LD 7 so had a sample too low for Labour and too high for the Conservatives . Their final in house weighting ( although not quite as bad as Comres ) produced an even more inaccurate result as they started with a sample that was not as accurate as the other pollsters . This was partly just bad luck as their penultimate poll was almost spot on before the in house weightings Con 41 Lab 39
|
|
thetop
Labour
[k4r]
Posts: 945
|
Post by thetop on Jul 6, 2017 10:57:13 GMT
Analysing the pollstera performance is not as simple as your table shows . Polling is a several stage process . A sample is taken , it is then weighted by demographic data to eliminate sampling errors and finally pollsters do additional weighting using in house assumptions on turnout and other factors . If you look at the final polls of the companies it gives a different story on accuracy . Examples Comres after initial weighting had Con 40 Lab 39 LD 9 so not that far out LDs too high , Con and Lab too low but gap pretty much spot on . They then introduced their own in house weighting and ended up with a massively erroneous forecast Ipsos Mori after initial weighting had Con 41 Lab 41 LD 7 pretty much spot on but after in house weightings they produced Con 44 Lab 36 LD 7 ICM after initial weightings had Con 44 Lab 38 LD 7 so had a sample too low for Labour and too high for the Conservatives . Their final in house weighting ( although not quite as bad as Comres ) produced an even more inaccurate result as they started with a sample that was not as accurate as the other pollsters . This was partly just bad luck as their penultimate poll was almost spot on before the in house weightings Con 41 Lab 39 That all may be true, but they don't give all stages equal credit - they hold the final figures up with the weighting they've devised as the one to use. I think it's perfectly reasonable to use those figures to determine their record. |
|
|
|
Post by bestkeptsecret on Aug 5, 2017 16:25:44 GMT
For me virtually all polls produce nonsensical results made all the more nonsensical by the way the outcomes are published
It seems that basic things like sample sizes go out of the window, telephone interviews are conducted by using a series of lead in questions to help get the answer the person conducting the poll thinks the publication commissioning it wants to hear, and the idea of representative samples seems to be completely ignored in many cases.
I took a look into how Survation (a company who conducts polls for the likes of the Mail on Sunday) conduct their polls. While they don't actually break any rules, they tread a fine line, and their methods seem questionable at best.
|
|
|
|
Post by Davıd Boothroyd on May 8, 2018 11:37:01 GMT
|
|
|
|
Post by archaeologist on May 8, 2018 19:03:56 GMT
So they now say there is a 90% chance the true figure is within 4 points of the polling figure. A poll of 40% then could be between 36 and 44%. Seems about time we should bury polls as well past their sell-by date! That level on [in]accuracy is useless.
|
|
Crimson King
Lib Dem
Be nice to each other and sing in tune
Posts: 9,861 
|
Post by Crimson King on May 8, 2018 22:26:27 GMT
to be fair it is not pollsters fault if politicians ignore the inevitable uncertainty in any sampling. Though treating them as useless is probably less stupid than treating them as accurate
|
|
|
|
Post by greenhert on May 11, 2018 16:24:00 GMT
So they now say there is a 90% chance the true figure is within 4 points of the polling figure. A poll of 40% then could be between 36 and 44%. Seems about time we should bury polls as well past their sell-by date! That level on [in]accuracy is useless. Polls are not past their sell-by date-the standards do need to be tightened, though, especially in the current political climate where Labour and the Conservatives are almost always close to each other in polling terms irrespective of whichever one of them is in the lead. |
|
|
|
Post by beastofbedfordshire on May 11, 2018 17:43:03 GMT
Polls are accurate when your party is ahead
|
|
|
|
Post by Davıd Boothroyd on Sept 28, 2018 11:35:37 GMT
BPC has decided on a new form of words over polling uncertainty:  |
|
|
|
Post by archaeologist on Jul 18, 2019 6:50:32 GMT
|
|
|
|
Post by justin124 on Jul 18, 2019 7:10:42 GMT
Interesting article - but it does not offer an explanation for Yougov's significant overestimate of the Brexit Party vote share at the EU election. |
|
Deleted
Deleted Member
Posts: 0
|
Post by Deleted on Jul 18, 2019 8:04:22 GMT
Interesting article - but it does not offer an explanation for Yougov's significant overestimate of the Brexit Party vote share at the EU election. or why ipsos which doesn't use false recall and got the euros right is closer to the polling average and yougov further away |
|
