Can anyone now say polls dont matter?

retrospooty

Platinum Member
Apr 3, 2002
2,031
74
86
Sure, one poll or another can be innacurate, but Realclearpolitics.com took all of them and gave us an average.

http://www.realclearpolitics.c...bama_vs_mccain/?map=10

When clicking on no toss up states... Their electoral map as of election day showed exactly what happened with one small difference. It had NC and IN going to McCain (just barely, by .4% for NC and 1.4% for IN. In actuality those 2 states are still too close to call as of 12:00 EST Nov 5th.

You cant get much more accurate than that.
 

EagleKeeper

Discussion Club Moderator<br>Elite Member
Staff member
Oct 30, 2000
42,591
5
0
The closer the competition, the more the polls can be inaccurate.
 

ProfJohn

Lifer
Jul 28, 2006
18,251
8
0
The polls had the out come right, but there are a lot of problem with their final numbers.

Look at the general election polls.

5 of them over estimated Obama's final total, only 2 over estimated McCain's. 5 of them underestimated Obama's total, while nine of them underestimated McCain's final total.
The good news is that 4 of them look to have gotten VERY close.

But some of them are WAY off. Gallup, Zogby, CBS, ABC, Marist all missed it by 3+ points.
 

retrospooty

Platinum Member
Apr 3, 2002
2,031
74
86
Originally posted by: ProfJohn
The polls had the out come right, but there are a lot of problem with their final numbers.

Look at the general election polls.

5 of them over estimated Obama's final total, only 2 over estimated McCain's. 5 of them underestimated Obama's total, while nine of them underestimated McCain's final total.
The good news is that 4 of them look to have gotten VERY close.

But some of them are WAY off. Gallup, Zogby, CBS, ABC, Marist all missed it by 3+ points.

Exactly... That's why I said "Sure, one poll or another can be innacurate, but Realclearpolitics.com took all of them and gave us an average. "

Instead of one poll with 1000 people, they are using over a dozen polls with 10's of thousands of people. Averaging the end result like RCP does turned out to be very accurate.
 

jonks

Lifer
Feb 7, 2005
13,918
20
81
Fivethirtyeight got it almost exactly right.

Predicted 6 point popular vote spread. Bam.
McCain (R) 46% 55,805,197
Obama (D) 52% 63,007,791
 

mshan

Diamond Member
Nov 16, 2004
7,868
0
71
Chuck Todd has previously said that the quality of publicly available polls is poor.

Do you think that is because of methodology (doubt it), or because they are trying to skew results towards a result they want (create narrative - how do they get paid and by whom?).

At the end of this campaign, it was very obvious that popular vote margin was going to be significant, as was electoral college, so there was no disincentive to actually to try and poll and report in an unbiased manner.

That being said, I suspect one would have to very carefully study polling track record over the course of the whole campaign to see who the really relatively unbiased pollsters were.

At the end, everyone is probably just jockeying for the next election cycle (look, we got it most right, even though it really didn't matter then).

 

Pens1566

Lifer
Oct 11, 2005
11,480
7,886
136
Originally posted by: jonks
Fivethirtyeight got it almost exactly right.

Predicted 6 point popular vote spread. Bam.
McCain (R) 46% 55,805,197
Obama (D) 52% 63,007,791

This ^
 

miketheidiot

Lifer
Sep 3, 2004
11,062
1
0
Originally posted by: mshan
Chuck Todd has previously said that the quality of publicly available polls is poor.

Do you think that is because of methodology (doubt it), or because they are trying to skew results towards a result they want (create narrative - how do they get paid and by whom?).

At the end of this campaign, it was very obvious that popular vote margin was going to be significant, as was electoral college, so there was no disincentive to actually to try and poll and report in an unbiased manner.

That being said, I suspect one would have to very carefully study polling track record over the course of the whole campaign to see who the really relatively unbiased pollsters were.

At the end, everyone is probably just jockeying for the next election cycle (look, we got it most right, even though it really didn't matter then).

the methodology was actually the problem, and this has been pretty widely discussed regarding cell phone issues, new registrations, etc.
 

mshan

Diamond Member
Nov 16, 2004
7,868
0
71
Let me clarify:

Statistical analysis is not the problem.

Bias on the part of the organization conducting the polls is the major problem. 3% margin of error is meaningless if construction and execution of study is deliberately biased to get certain desired results. A p-value < 0.05 again can be meaningless if study is intentionally or even unintentionally biased by conductor of study. If there were no external oversight from outsiders saying your study is flawed, would you trust a drug company paid for study, done by researchers working for that drug company, that said a known dangerous drug was safe because of a study they produced for you said so?

Typically, they aren't just trying to report the "news"; they are trying to create, or at least direct it in a desired direction.


 

ProfJohn

Lifer
Jul 28, 2006
18,251
8
0
Originally posted by: miketheidiot
the methodology was actually the problem, and this has been pretty widely discussed regarding cell phone issues, new registrations, etc.
The problem with that argument is that the people pushing it all suggested that Obama should be doing better than the polls because of those problems.

But that wasn't the case. In general the polls under estimated McCain's support and over estimated Obama's.

As some have pointed out, this is a problem mainly with the how they skew the outcome. The voter model. How many D's vs. R's in each poll. The polls that had that balance wrong were WAY off, the polls that had that number right were close.

But the whole cell phone thing is blown out of proportion. Didn't someone suggest that Obama was being under polled by 2 points due to that problem? And yet at the end he ends up almost 2 points below the RCP average. That cell phone guy would have been wrong by 4 points, huge!
 

fskimospy

Elite Member
Mar 10, 2006
83,717
47,406
136
Originally posted by: mshan
Chuck Todd has previously said that the quality of publicly available polls is poor.

Do you think that is because of methodology (doubt it), or because they are trying to skew results towards a result they want (create narrative - how do they get paid and by whom?).

At the end of this campaign, it was very obvious that popular vote margin was going to be significant, as was electoral college, so there was no disincentive to actually to try and poll and report in an unbiased manner.

That being said, I suspect one would have to very carefully study polling track record over the course of the whole campaign to see who the really relatively unbiased pollsters were.

At the end, everyone is probably just jockeying for the next election cycle (look, we got it most right, even though it really didn't matter then).

I find it highly unlikely that any polling firm was attempting to be biased. These pollsters do tons of polling for all sorts of things in non election years, and nobody wants to have the reputation as the guy who got it wrong. It directly impacts their bottom line.

The polling itself is highly accurate in determining preferences, but the thing is that voting is not only a revealed preference it is an action taken. That's where pollsters run into trouble, because they are all wrestling with how they will guess voter turnout will be. That's what causes errors in polling.

Polling very much matters, why do you think campaigns spend so much money on it? They aren't dumb, they know what works.
 

miketheidiot

Lifer
Sep 3, 2004
11,062
1
0
Originally posted by: mshan
Let me clarify:

Statistical analysis is not the problem.

Bias on the part of the organization conducting the polls is the major problem. 3% margin of error is meaningless if construction and execution of study is deliberately biased to get certain desired results. A p-value < 0.05 again can be meaningless if study is intentionally or even unintentionally biased by conductor of study. If there were no external oversight from outsiders saying your study is flawed, would you trust a drug company paid for study, done by researchers working for that drug company, that said a known dangerous drug was safe because of a study they produced for you said so?

Typically, they aren't just trying to report the "news"; they are trying to create, or at least direct it in a desired direction.

statistical analysis and methodology was the problem.
 

miketheidiot

Lifer
Sep 3, 2004
11,062
1
0
Originally posted by: ProfJohn
Originally posted by: miketheidiot
the methodology was actually the problem, and this has been pretty widely discussed regarding cell phone issues, new registrations, etc.
The problem with that argument is that the people pushing it all suggested that Obama should be doing better than the polls because of those problems.

But that wasn't the case. In general the polls under estimated McCain's support and over estimated Obama's.

As some have pointed out, this is a problem mainly with the how they skew the outcome. The voter model. How many D's vs. R's in each poll. The polls that had that balance wrong were WAY off, the polls that had that number right were close.

But the whole cell phone thing is blown out of proportion. Didn't someone suggest that Obama was being under polled by 2 points due to that problem? And yet at the end he ends up almost 2 points below the RCP average. That cell phone guy would have been wrong by 4 points, huge!

actually, if you divide the polls into 'cell phone' and 'no cell phone' polls, the 'no cell phone' polls missed it by about 2-3 points, and the 'cell-phone' polls missed it by about 3-5% And he hit the rcp average right on the money btw
 

mshan

Diamond Member
Nov 16, 2004
7,868
0
71
Obama's whole strategy was built around the electoral college, just like his strategy in the Democratic primaries was built around getting the required number of delegates (he had said that if the Democratic Primary had been about popular vote, he could have run up his totals in large states like Illinois).

For what it is worth (not much) I predicted 52% - 45% months ago based upon two simple observations: economic models predicted Obama victory in 52 - 55% range this summer and average of all polls seemed to indicate McLame always was bumping up against that 45% poll of polls ceiling.

Most polls were either 51% or 52% for Obama and 45% or 46% for McLame. Meaningless differences in the macro numbers, in the end...

McLame's macro numbers may have tightened up with rally in stock market, but I suspect that damage in state by state electoral votes persisted.

This and this (obviously ones for the presidential election, not Democratic Primary) are the immediate pre-election data I would really be curious to see.

 

retrospooty

Platinum Member
Apr 3, 2002
2,031
74
86
Originally posted by: ProfJohn

But that wasn't the case. In general the polls under estimated McCain's support and over estimated Obama's.

!

What?

The whole thing discussed in this thread is how close to reality the polls were. The olny thing that was slightly off is that they said NC and Indiana would be won by McCain (barely) . In reality they are still too close to call.
 

First

Lifer
Jun 3, 2002
10,518
271
136
Yup, RealClearPolitics got it exactly right, with Missouri being just as close as they predicted it would be (i.e. tied). Good to see that reliable, aggregate polling numbers are once again proven superb indicators of election results. McCain got a very fair shake in the national and state polls.