2012 presidential polling errors

[ceejay]

Thinking back a week, it seems to me that the final polling numbers seemed to systematically overstate support for Romney. For the purposes of discussion, I am just looking at national polls. Depending on which aggravation you look at these differences vary. For example, http://www.realclearpolitics.com/epolls/2012/president/us/general_election_romney_vs_obama-1171.htm shows an error of 2.0% towards Romney. The 538 aggregation shows that most polls were biased (statistically, not necessarily ideologically) for Romney as well.

Because aggregation of multiple polls effectively increases sample size and reduces so-called margin of error, there must have been similar systematic (although not necessarily large) errors. I am just wondering what other people think could cause this. I am not looking for a partisan discussion here, but rather a discussion of analytical techniques.

I think correction of polling data to account for cell phone limitations/restrictions is too simplistic. I would expect that any responsible poll would at least attempt to correct for that. I would look into the "secret sauces" which make up the likely voter models. Additionally, I would look into under representation of Hispanic vote by surveys that do not allow responses in Spanish.

[Ocala Mike]

[QUOTE=ceejay]

Additionally, I would look into under representation of Hispanic vote by surveys that do not allow responses in Spanish.[

/QUOTE]


Bingo!

[Tom]

In other words, bad polling practices.

[SchagFactorToWin]

I'm not convinced that combing polls results in a lower margin of error. But even if it does, the design error rate increases, possibly exponentially. If a poll has some design error, as most do, that error rate would grow when combined with another poll.

Use handicapping as an example. Let's say I have a system that shows long shots come in more often with a full moon. You have a system that shows long shots come in more often on Tuesdays. So, if we combine them, looking for long shots on Tuesdays with a full moon should be great! The problem is that both systems are flawed (in this case, a correlation/causation error), bringing about a worse predictor than a better one.

[Tom]

First off, I love your avatar! :D
Secondly, I think we have to many polls.
Polls should reflect public opinion, not shape it.

Imagine an election where polls and political contributions were both outlawed.

And TV commercials.
And radio ads.

[lamboguy]

not that this a poll, but i thing thegreek.com was very accurate based on their odds. i think their data was a bit more scientific than a regular ordinary poll such as Gallop or Rasmussen,

[Steve R]

Thinking back a week, it seems to me that the final polling numbers seemed to systematically overstate support for Romney. For the purposes of discussion, I am just looking at national polls. Depending on which aggravation you look at these differences vary. For example, http://www.realclearpolitics.com/epolls/2012/president/us/general_election_romney_vs_obama-1171.htm shows an error of 2.0% towards Romney. The 538 aggregation shows that most polls were biased (statistically, not necessarily ideologically) for Romney as well.

Because aggregation of multiple polls effectively increases sample size and reduces so-called margin of error, there must have been similar systematic (although not necessarily large) errors. I am just wondering what other people think could cause this. I am not looking for a partisan discussion here, but rather a discussion of analytical techniques.

I think correction of polling data to account for cell phone limitations/restrictions is too simplistic. I would expect that any responsible poll would at least attempt to correct for that. I would look into the "secret sauces" which make up the likely voter models. Additionally, I would look into under representation of Hispanic vote by surveys that do not allow responses in Spanish.
I don't see an issue here at all. In reviewing the final RCP state polls the only one of 50 that was wrong was Florida, which they listed as a tossup anyway.

[lsbets]

Every major poll was within the margin of error. The difference on Election Day was the minority percentage of the vote was higher than it was in 2008. Nobody projected that except for the Obama campaign. Did the pollsters get their models wrong? Yes. The question is why.

My theory, which I stated weeks before the election, is the Obama campaign wanted to make this a base election. The wanted undecideds to stay home. Assuming those undecideds were white voters, and that they were so turned off by both candidates they decided not to vote, the Obama strategy worked perfectly. Undecided white voters came out in small numbers, base minority voters came out in large numbers. When the pollsters go back and apply the actual turnout to the results instead of their projections, the polls were dead on.

[PaceAdvantage]

The polls probably didn't account for the "maddog nonexistent" voting fraud... :lol:

[ceejay]

Polls should reflect public opinion, not shape it.

Tom,

Are you asserting some sort of observer effect or uncertainty principle? That the mere measurement process of the poll affects the outcome of the poll and/or the election?

[Tom]

I'm suggesting that that is the only thing polls do.

http://www.charlotteobserver.com/2012/09/15/3532683/polling-methods-or-bias-can-taint.html

“Polls aren’t just a snapshot of public opinion anymore, they’re advocacy or even propaganda and often come with an agenda,” says Chris Fitzsimon, executive director of the liberal N.C. Policy Watch.
Read more here: http://www.charlotteobserver.com/2012/09/15/3532683/polling-methods-or-bias-can-taint.html#storylink=cpy

[horses4courses]

Imagine an election where polls and political contributions were both outlawed.
And TV commercials.
And radio ads.

You mean like North Korea? ;)

I do agree with you, though.
The polling has reached the overkill stage, and the money spent on campaigns is outrageous.

[Tom]

N Korea only has one candidate! :D

[Ocala Mike]

Tom,

Are you asserting some sort of observer effect or uncertainty principle? That the mere measurement process of the poll affects the outcome of the poll and/or the election?



I love it, ceejay! Quantum Theory as applied to election polling. :ThmbUp: :ThmbUp: :ThmbUp: :ThmbUp: