So in case anyone was still in doubt about this, it turns that polling — real polling by independent organizations, not a campaign’s leaked internals — works. It actually works pretty darn well.
We noted after the primary that the polls correctly picked up a big late surge in support for Deval Patrick, and the ones taken within a few days of the primary were within a couple of points of correctly predicting the outcome.
Same thing happened in the general election. All the polls taken within the last two weeks or so were close, and the ones taken within a week of the election were almost exactly right. Seizing on the clever suggestion of stomv and alexwill, I’ve normalized this walk down memory lane of poll numbers by dividing each reported number by (1-%undecided/refused), so they don’t match exactly the numbers reported in the original polls, but they’re a better guess at what the polls would have shown if you forced the undecideds to choose.
ACTUAL (11/7): DP 56%, KH 35%, CM 7%, GR 2%.
Suffolk (11/2-5): DP 58%, KH 34%, CM 7%, GR 2%.
SHNS (11/1-2): DP 57%, KH 30%, CM 9%, GR 2%.
SUSA (10/30-11/1): DP 57%, KH 35%, CM 6%, GR 3%.
I could go on, but you get the idea. So the next time a campaign spokesguy or sympathizer tries to downplay polls showing his candidate way down by saying that the campaign’s internals show a much tighter race, smile, nod politely, and go about your business.
NH-02, the public polls had been all over the place for months. The final one, the day before the election, was pretty accurate – Hodes won with 53% to Bass’ 45% and I think that’s about what WMUR/UNH predicted. But the previous 2 were one from Research 2000 that had the race tied at Bass 47% – Hodes 46%, and the Constituent Dynamics one that had it at Hodes 50% – Bass 47%.
<
p>
And looking back at older polls leaves us with the choice of believing the internal polls that showed the race just about tied in late August, and that Hodes capitalized on that, or that Hodes was trailing by like 20 points less than 2 months ago and turned it around by that much, that quickly.
<
p>
Meanwhile, in NH-01 no polls were predicting a Shea-Porter victory.
<
p>
So I’m inclined to think that the issue is at a minimum more complicated. And as well, I’d note that in a comment at Daily Kos, Charlie Cook said he relied more on internal-type polls than a lot of these public ones.
Remember Tim Murray’s LG primary race? I believe the polls had tapped Goldberg as the front-runner rather consistently, right up to the race. Murray’s internals indicated these polls were wrong, and the actual vote corroborated this.
though I think one can distinguish the LG’s race on the ground that almost no one was paying attention to it — undecideds remained absurdly high right up to the last day, IIRC. For races that people are really paying attention to, like the Gov’s race, I think one disregards public polling at one’s peril.
In fact, didn’t “Undecided” lead all candidates in every poll for that race, up to the end?
Patrick’s outstanding GOTV operation was supposed to confer an advantage of anywhere from three to five points. But as you point out the polls essenbtially called it.
<
p>
Either ther’s no ground advantage, or it was cancelled by racism or other factors hidden from pollsters, or the polls just happened to be wrong by exactly the margin the GOTV work provided.
<
p>
Absent exit polling we’ll never know.
It is hard to deny that Patrick’s ground game was a large as any seen before, but I agree you would have thought it would translate into a lead larger than the polls indicated.
These are polls of likely voters, if I recall correctly. What this tells me was that the polls had the likely voter model right, not necessarily that the Patrick ground game didn’t matter.
I’m trying to understand why a poll of likely voters matched very closley with the actual result.
<
p>
This would seem to suggest that the GOTV of both parties was nearly equal. Which I find hard to believe.
<
p>
Any thoughts?
is that the polling might have accurately picked up the impact of Patrick’s ground game. That is, folks who in other years wouldn’t have been identified as “likely voters” were identified as such this year by pollsters, because the Patrick operation had already convinced them that they should get to the polls this time.
<
p>
Right, sco?
There’s not question, though, that turnout among Dems was key to Patrick’s victory. He won because Democrats came out for him who didn’t come out for O’Brien or Harshbarger. If the polls captured that, it’s because the polls were good, not because the turnout didn’t matter.
You’re absolutely right. It was amazing. In fact, the Channel 7 pollster predicted Deval by 22%! Amazing.
So the next time a campaign spokesguy or sympathizer tries to downplay polls showing his candidate way down by saying that the campaign’s internals show a much tighter race, smile, nod politely, and go about your business.
<
p>
Unless it is YOUR Candidate who is down in the polls of course. In that instance YOU will be the one downplaying the polls. You’ve got to believe in your candidate and fight the perception of the polls for those voters who want to vote for the winner who is leading in the polls. Yeah the polls were in our favor this time, but if they weren’t….and they won’t always be….we’d be doing the same thing.
that we’d be doing it, but the point is that we’d be fibbing — just like the folks claiming that Healey was within single digits were fibbing.
<
p>
Also, it’s important to note that the only time you should really worry TOO much about horserace numbers is within a week or two of the election. As we know, polls are only a snapshot at that moment in time. My point is that they are generally a very accurate snapshot. But as time goes on, the picture can change. The less time there is ’til the election, though, the less likely that is to happen.
As has been discussed earlier, there has been a tendency in the past for polls to overestimate support for black candidates. That does not seem to have held true in 2006. The last polls in Tennessee seem to have been accurate, showing Ford trailing by 3-5 points — he lost by 3.
<
p>
Michael Steele did poll a little closer in Maryland than the 10 points he lost by. You might be able to make a case that black Republicans are affected more than Democrats. But it doesn’t seem to have been too significant there. And Ken Blackwell in Ohio appears to have had the crap beat out of him in the election by about the same amount he had the crap beat out of him in the surveys.
<
p> – Dan