UPDATED: Ugh, of course you are right david, my mistake and I have updated the popular vote to the ‘current’ number (could be as many as 7 million votes still out there?). Classic math blunder, I used the totals from the official Google results, which didn’t mention the third (and fourth) party candidates and I forgot to look it up myself. Thanks for the heads up.
Wow. What a night last night, and not just for Democrats, Massachusetts, gay rights, and women all over the country (bye bye Akin, hello record representation).
As was discussed here at BMG, something else was apparently at stake: the use of numbers, averages, and basic math as a way to determine relative counts of things. And basically, it was a huge night for geeks — not just my personal favorite Nate Silver, but also other aggregators (RealClearPolitics for all their conservative editorializing had all but Florida right in their final state by state poll aggregations) and other election models based on polls (Princeton Election Consortium also called all 50 states too, with Florida as a tossup slightly leaning Obama). The victory of numbers over pundits was the icing on the cake for me.
An overview of the charts below: I’d call pablo and patrick the BMG winners this time around, though with the updated correct numbers the winners are a bit more diverse. They both called the Warren/Brown race to under a point of total deviation and nailed the electoral college exactly. pablo was one point away from the Obama popular vote and although patrick didn’t participate in that one, he DID nail the Tierney surprise victory as well (for a bonus). sabutai got VERY close on the popular vote prediction.
Anyway, the leader board for the BMG Predictions (as measured by standard deviations for the predictions):
1) Warren / Brown popular vote percentages
User patrick appears to be our closest prediction, just .6 total points away from the final Warren Brown numbers, with pablo, fenway49, and trickle-up right behind. The election favored the bold prediction, with Warren outperforming the average BMG prediction. The 538 model was about middle of the pack, like the BMG Average.
2) Obama / Romney Popular Vote
UPDATED, courtesy of david’s correction. New winners are: sabutai, alexwill, and chisinandover. Sweet, I crack the top 5 now. 538 projection very close on the popular vote, also makes the top 5.
User whose night still fell wildly short of expectations: Bob_Neer.
3) Obama / Romney Electoral College Votes
Three users nailed the Electoral College vote: eunomia, pablo, and patrick. The BMG average and the 538 model were closest besides that (the 538 model, extrapolated to each state percentages, would have matched the exact total, but they accounted for the odds rather than a straight binary prediction).
Nice job everyone! Thanks for participating in this diversion and my little experiment!
BTW, my own attempt at a predictive model (Model D above) did very poorly. It was based on adjusting the predictions for bias, but it’s pretty clear to me now that’s probably not a great basis for a model. Although it got the Brown victory right last time, it went the wrong direction this time and underestimated Warren’s win.