BMG prediction averages:
Turnout: 399,927 (12%)
So BMG predictions average out to a 37-32 Coakley win over Capuano, with Khazei in distant third.
I think because they are a collection of predictions rather than a poll, these numbers represent the Conventional Wisdom, but with a BMG slant. To figure out which direction these numbers slant, we would need to look at straw polls within the BMG community. BMG Straw Poll
BMG Straw poll results:
So the BMG CW slants towards Capuano, with Khazei much more popular than Pags. To find the likely general population CW, i think we would have to adjust Cap and khazei down a bit and Coakley and pags up a bit (-4 points for cap, +3 for coakley, -3 for Khazei, +4 for pags). So the actual general population conventional wisdom is probably closer to
There is a lot of guesswork there as to what the effect of the bias is, but that feels about right to me. People think Coakley has a twelve point lead over Capuano and Pagliuca and Khazei are a tossup for third.
But these predictions only give the Conventional Wisdom, and I think this time the CW is off. I think Coakley’s support has been steadily dropping as the date gets closer and people will look at the latest polls and realize that Capuano has the best chance. He will peel off
from voters who left Coakley to go to Pags and Khazei, and I’ll be surprised if she breaks 35%.
With the institutional support of the mayors and congressional delegation, Capuano also has a local organizing advantage.
Given all that, MY final prediction is:
So I think Capuano wins tonight in a squeaker! But if CW is right, he loses to Coakley 40 to 28. This means if he does win it will be seen as a big upset in the media.
Thanks for all the hard work, and I’d say this is the most useful first post I’ve seen in a while. Usually we open accounts to yell at people.
p>Cead Mille Failte to BMG!
you did not make your prediction until you had the benefit of the conventional wisdom of BMGers…so you had a decided edge.
p>I don’t think you are right, but time will certainly tell.
i’m not sure i necessarily followed the collective wisdom of BMG, but i definitely did wait long enough to take advantage of it. although my actual personal prediction is based mostly on hunches and what i’ve seen of their campaign efforts.
p>i’m actually more curious to see how close the BMG composite is to the final tally…
Thanks for the number crunch
as long as you are kind of steady and slow with your mouse:
have a reasonable idea of the map of Massachusetts.
if I had stuck to my first prediction I would have been quite close to the final outcome. I think I had Coakley at 45, Capuano at 28, Pags at 15 and Khazei at 12 at one point, a lot closer than my prediction listed here.
p>Oh well its quite unfortunate, but I’d have been more upset if Khazei’s votes would have been enough to elect Capuano.
p>The results are rather unfortunate, I guess I am now an undecided voter going into the general election (not between Coakley and Brown but between whether I will vote or not)
that looks way better. thanks for making it readable.
Sleeples give laurels to Petr and Hoyapaul for their high degree of partial accuracy. But how do you weigh overall accuracy?
p>No single answer to that, of course, but one time-honored method is by total deviations.
p>For example Petr was off by only 1 for Cap and 2 for Coakley, but 8 for Khaz and 6 for Pags, for a total score of 17. Hoyapaul was more accurate at 14.
p>That’s not bad, but the best overall was 12, scored by GraySky and Bostonboomer.
p>Here’s the damage:
30 Steve Stein
33 Peter Porcupine
p>Of course that does not account for the turnout question.
p>Of interest: “Groupthink,” the composit average calculated by sleeples, did relatively poorly at 30.
It is easy to read, understand, and rank. although I don’t know how you’d add turnout into there, except maybe as a tie-breaker? Separate category?
p>I think the originators of the predictions will have to decide how they judge closeness, but I would put in a vote for this method. Seems fair to me.
I haven’t checked all of them yet, but I wonder how many people got the order of the four candidates correct? I was solid on Coakley and Capuano, but had Pags and Khazei flipped and was pretty far off on their percentages (prediction of 16% for Pags and 8% Khazei).
p>Robbed! Robbed I was!
p>All seriousness aside: there ought to be a penalty for getting the order wrong, for instance GraySky called Pagliuca over Khazei.
p>And if you were really serious about it, you’d not assign one point per percentage point off, but rather do something like (treating the actual results as the distribution) assigning one point per standard deviation off, mixing in number of standard deviations from predicted turnout as a multiplier. This would weight accuracy more accurately (and I would win…=-) I’m not serious enough about it to do that, but it would be more fair.