Sooo... what happened in the 2008 Democratic NH primary?

[Joe Republic]

I never really got a definitive explanation for this:

January 1st Poll Average: Clinton +17
January 2nd Poll Average: Clinton +16
January 3rd Poll Average: Clinton +7
Iowa Caucus Result: Obama +5 on Clinton
January 4th Poll Average: Obama +0.25
January 5th Poll Average: Obama +2
January 6th Poll Average: Obama +8
January 7th Poll Average: Obama +9

Final Primary Result: Clinton +3


The most frequent explanations given are the Bradley Effect and Clinton's last-minute tear-jerking incident, but I still find it a little hard to believe that they can account for such a massive disparity between every poll released in the last couple of days and the actual result.

What is your view?

[Joe Republic]

Supporters of other candidates felt that a long race could possibly benefit them, mainly Edwards and Richardson.  They feared that if Obama won New Hampshire it would be over (similar to Richardson's supporters giving Obama Iowa).  Edwards and Richardson both fell a few points short of what they were polling, and Hillary was the one that these points went to.

The median average of polls for the 6th and 7th of January was as follows:  Obama 38%; Clinton 29%; Edwards 19%; Richardson 6%; Others 3%.

The actual result was as follows: Obama 36%; Clinton 39%; Edwards 17%; Richardson 5%; Others 3%.

The difference in polling and results for Clinton's and Obama's competitors is clearly not enough to account for such a difference overall.

[Joe Republic]

Well, the reason I got a different spread was because I discounted some of the polls.  For example, the Zogby (always an awful poll) which had Obama up 13% and the Uni polls that came out.  Rasmussen for example had the polling at 37-30-19-8, when the actual result was 36-39-17-5

Which comes out to:
Obama -1
Clinton +9
Edwards -2
Richardson -3

To mean, that means undecideds broke late to Clinton hard, as did supporters of other candidates.  As you can see, Obama got what he polled.

For this type of analysis a median average is preferable to a mean average.  It excludes outliers by definition.

The undecideds breaking overwhelmingly for Clinton is a credible theory, but the median average of that group in the last two days was 5%.  Even if all of them and a contingent of the other candidates' supporters swung to Clinton, we've still only accounted for 8% to add onto Clinton's polling, when in fact she performed better by 12%.

[Joe Republic]

They also chronically overestimated the youth vote and underestimated the impact of blue collar voters in the cities and along the MA border.

This is also a credible theory, but I'm extremely surprised that eleven different polling agencies who released polls in the last two days could have made such a misjudgement.  Especially since NH has been a polling hotspot for decades.

[Joe Republic]

Joe, after reading this over again, I'm thinking there is obviously something wrong a poll swing of 10 points in one day (January 2-3) w/o any voting, and a 26 pt swing in a week regardless

The daily 'averages' I cited for the 1st and 2nd of January were in fact made from just one poll each day - sorry if that seems a little misleading.  Anyhow, those polls were both by Suffolk University, which as well having the faults you'd expect for a university pollster, also got their final result wrong.

As for the 26-point swing in a week (which I highly doubt was realistically that large, but still), don't underestimate the impact the Iowa caucus had.  It was the first real game changer, and surely you must remember that the media first began to speak of Obama's inevitability after that point.  New Hampshire was hugely affected.