Nov 2020
8:28pm, 17 Nov 2020
12,271 posts
|
rf_fozzy
A poll that predicts the correct winner, but is 10% off (i.e. candidate x wins by 20% rather than 20%) is more wrong than a poll that says candidate y will win by 2%, but then candidate x wins by 2%. The former *should* be outside the MoE, the latter will be within the MoE. The polls are *not* an accurate black/white measure. We have to learn to understand that they actually on project a probability distribution. That's why Joe Biden had a 90% chance of winning (and not much higher), because it projected that he could withstand what actually happened, which was a standard polling error in most states towards Trump. The same is true for the Clinton/Trump in 2016. Clinton had a greater chance of winning *based on the available data* but the 1/3 chance that was given to Trump recognised the fact that all it needed was a normal sized polling error in Trump's direction. Which is what happened. And although there *does* seem to be a slight issue with Trump and polls (no, it's *not* shy trump voters - there's no evidence in the data for that), no-one mentions that in 2008 and 2012, there was a polling error that underestimated Obama. Just because he won, doesn't mean it wasn't there. |
|
Nov 2020
8:29pm, 17 Nov 2020
12,272 posts
|
rf_fozzy
Sorry, typo. First bit should have said: "A poll that predicts the correct winner, but is 10% off (i.e. candidate x wins by 20% rather than 10% as stated by the pre-election poll) is more wrong than a poll that says candidate y will win by 2%, but then candidate x wins by 2%. |
Nov 2020
8:49pm, 17 Nov 2020
3,236 posts
|
Pothunter
Can’t remember if I said it here before so apologies if I’m repeating myself! I was talking to my dad in Australia a couple of weeks ago and he mentioned an article that suggested instead of a pollster asking who you intend to vote for they should ask who you think your neighbour would vote for. Might get a more accurate view of intentions as someone would doesn’t want to admit to someone on the phone that they were planning on voting for Trump/Brexit/Tories would happily give an opinion on their neighbour’s intent |
Nov 2020
8:53pm, 17 Nov 2020
37,630 posts
|
Nellers
I don't think that's the issue though Cheg. It's more getting a sample of people who will respond to the pollster at all.
|
Nov 2020
8:59pm, 17 Nov 2020
2,071 posts
|
Cheg
95% confidence. 60% accurate. newsroom.haas.berkeley.edu |
Nov 2020
9:14pm, 17 Nov 2020
13,221 posts
|
NDWDave
In the US election, 60/40 is absolute obliteration. This year it is looking like 51/47 in the popular vote. 50/49 would have been a repeat of 2016 whereas 52/46 would be a “landslide”. Not huge margins in the statistical world
|
|
Nov 2020
9:18pm, 17 Nov 2020
12,273 posts
|
rf_fozzy
And of course the polling of the popular vote in 2016 was bang on. The issue is that the popular vote is not relevant for the electoral college. And the MoE on state polling is necessarily larger. |
Nov 2020
9:26pm, 17 Nov 2020
24,102 posts
|
Johnny Blaze
The last few Wisconsin polls had Biden ahead by between 8 and 13 percent. He won by 0.7%. Seems to me the polling problem is pretty deep-seated.
|
|
Nov 2020
9:32pm, 17 Nov 2020
12,274 posts
|
rf_fozzy
Well for starters, don't look at individual polls - look at the polling average ("wisdom of crowds" en.wikipedia.org ) as that's more likely to be a better prediction. But yes, sorry I forgot Wisconsin (as well as Florida) does seem to have a specific issue. One plausible explanation seems to be the Covid outbreak in WI was (and still is) particularly bad on election day. But we don't have the full data to explain it yet. However, MI, PA, MN and yes, even IA were all within the MoE Remember that on state polling the MoE is about 3% - and that's for each candidate, so a 6% shift from poll to result is within the MoE |
|
Nov 2020
9:38pm, 17 Nov 2020
12,275 posts
|
rf_fozzy
So currently trying to understand why "purple pen" won't export files and I get a "a generic error occurred in gdi+" and it doesn't seem straightforward. So I don't have much time to argue now. So I'll point you in the direction of Nate Silver: fivethirtyeight.com and the podcast they did about it here: fivethirtyeight.com |
Useful Links
FE accepts no responsibility for external links. Or anything, really.Related Threads
-
Fantasy General Election
Jul 2024
-
EU Referendum - In or Out? Vote here
Aug 2018
-
March to Parliament Against Brexit - Sat 2nd July
Jun 2016
-
EU Referendum
Feb 2016
-
Ads on Fetch - anyone else getting Leave and Remain?!
Feb 2017
-
The Environment Thread :-)
Dec 2024
-
Economics
Aug 2023
-
Dear Scottish Fetchies
Jan 2023
-
Any economists out there - question
Oct 2022
-
Power and exploitation - please check my sanity
Oct 2018