[darjeeling]

Here's a fuller dataset from http://electoral-vote.com/

102 swing state polls within final 2 weeks from pollsters producing at least 4 polls. ARG, Rasmussen, Gravis and Pulse Opinion all showed a pro-Republican bias. Zogby, SurveyUSA, PPP, and ORC Intl showed no bias. Non-swing state polling was a slightly different story, but not of much importance in predicting who was going to win.

As others have said, the aggregators did indeed do a remarkable job of predicting the outcome, successfully correcting for biases where they occurred. They also got well ahead of the columnists in identifying when momentum shifted in the race; Republican spinners were still claiming momentum for Romney almost two weeks after the poll aggregators showed he was going backwards.

Pollster	Polls	Mean polling error (MPE) in points	MPE standard error	prob. no bias
ARG	4	REP+4.5	1.5	0.06
Rasmussen	14	REP+4.2	0.6	6E-06
Gravis Marketing	13	REP+2.6	0.5	2E-04
Pulse Opinion Research	5	REP+2.5	0.9	0.06
Marist Coll.	8	REP+1.5	1.1	0.22
IPSOS	12	REP+1.1	0.8	0.19
PPP	27	REP+0.3	0.4	0.45
ORC International	4	REP+0.3	1.1	0.82
SurveyUSA	8	REP+0.2	0.9	0.82
Zogby	7	REP+0.1	1.7	0.93

[Raphael]

It's a little interesting that there are no pollsters that had a democratic bias, and all the unbiased ones had a slight republican leaning. I'm not trying to conspiracy theorize - it's more likely to be related to old fashioned methodologies targeting old fashioned voters better (who tend to be republican voters) - but it begs the question of whether this shows that pollsters need to get their act together and try to find a way to target young people.

(It's also possible that the flaw is caused by another demographic split, mind you. THe age one just seems more likely, since it's pretty likely that's why Rasmussen were so off)

[darjeeling]

Originally Posted by Raphael

It's a little interesting that there are no pollsters that had a democratic bias, and all the unbiased ones had a slight republican leaning. I'm not trying to conspiracy theorize - it's more likely to be related to old fashioned methodologies targeting old fashioned voters better (who tend to be republican voters) - but it begs the question of whether this shows that pollsters need to get their act together and try to find a way to target young people.

(It's also possible that the flaw is caused by another demographic split, mind you. THe age one just seems more likely, since it's pretty likely that's why Rasmussen were so off)

The slight bias in the main group of pollsters could actually be due to Obama picking up a few extra voters in the closing days, perhaps people who approved of his handling of the response to storm Sandy. Outside the swing states, Obama did out-perform his polls in North-Eastern states in the path of the storm, notably New Jersey.

There was a clever analysis of the different polling companies by Drew Linzer of votamatic.org, showing they essentially split into two groups - a small number of multi-polling firms (Rasmussen, ARG and Gravis) favouring Romney, and a much bigger group of the rest of the major firms plus all the one-off pollsters, who collectively scored the race around 1.5 points more in Obama's favour (link). While there were some small house effects amongst the main group, there weren't any that seemed to show a
pro-Obama bias anything like the pro-Romney bias of Rasmussen et al.

The actual results now show - unsurprisingly - that the mainstream pollsters were right.

[LuckyLloyd]

Originally Posted by Black Swan

Both Rasmussen and Gallup polls favoured Romney during the past month up to the election. They will have to rethink their sampling methods before 2014 midterms, or potentially suffer the same fate again with Congressional races.

Silver gained a lot of credibility in this election. His prediction model had been superior to Rasmussen and Gallup.

That's quite the understatement! He nailed the Presidential election to the wall.

[Ozymandius2011]

Originally Posted by darjeeling

Here's a fuller dataset from http://electoral-vote.com/

102 swing state polls within final 2 weeks from pollsters producing at least 4 polls. ARG, Rasmussen, Gravis and Pulse Opinion all showed a pro-Republican bias. Zogby, SurveyUSA, PPP, and ORC Intl showed no bias. Non-swing state polling was a slightly different story, but not of much importance in predicting who was going to win.

As others have said, the aggregators did indeed do a remarkable job of predicting the outcome, successfully correcting for biases where they occurred. They also got well ahead of the columnists in identifying when momentum shifted in the race; Republican spinners were still claiming momentum for Romney almost two weeks after the poll aggregators showed he was going backwards.

A surprisingly good election year for Zogby, hitherot referred to by Nate Silver as "the worst pollster in the world".

[darjeeling]

Originally Posted by Ozymandius2011

A surprisingly good election year for Zogby, hitherot referred to by Nate Silver as "the worst pollster in the world".

Zogby's polls (on-line rather than phone polls, hence the criticisms) had much bigger margins of error than the PPP and other firms' polls, but not the systematic bias seen for Rasmussen.

The Zogby polls up on the Electoral Vote site seem to be only a snapshot of the full tracking polls (not sure why that is), and neither RCP nor Nate Silver used them at all. Silver did though use the Gravis polls, which Electoral Vote chose not to use.

Very complicated, this polling business!

[nagilum2]

Originally Posted by darjeeling

Zogby's polls (on-line rather than phone polls, hence the criticisms) had much bigger margins of error than the PPP and other firms' polls, but not the systematic bias seen for Rasmussen.

The Zogby polls up on the Electoral Vote site seem to be only a snapshot of the full tracking polls (not sure why that is), and neither RCP nor Nate Silver used them at all. Silver did though use the Gravis polls, which Electoral Vote chose not to use.

Very complicated, this polling business!

I thought Silver would nail everything except FL and VA, maybe CO. He definitely earned my respect in this cycle (and I already respected his number crunching abilities a lot).

[Ozymandius2011]

The SC is due to hear a challenge to the part of the 1965 Voting Rights Act requiring mostly Southern states to get permission from the DOJ before changing electoral-laws. Chief Justice John Roberts (of Texas) has expressed concerns it may be outdated, despite the fact that the Bush DOJ had to intervene when Shelby County, Alabama violated the Act in 2005. Not surprisingly, that is the county making the challenge.

There were reports in Alabama that road-blocks were placed around a polling-station in an African-American area.