The worst prediction in terms of margin was Hawaii (the topmost dot), where two polls led me to guess an Obama margin of 27 points that was actually 15.8 points too Republican.
Here's the same graph zoomed in on the battleground states.
My prediction for every battleground state except Ohio was too Republican. The median poll spread method I chose called Michigan a 3 point state (ludicrously). The real margin was 8.5 points. Iowa, Colorado, New Hampshire, Nevada -- all these states went more than 3.5 points more for Obama than I had predicted using the median poll spread. The model got the call on Florida wrong, and it predicted only a tie for Colorado, which wasn't even the next closest state (Ohio was).
On average, my guesses had a 1.6 point Republican house effect overall, but a 3.3 point Republican shift in states that went for Barack Obama.
The worst prediction in terms of margin was Hawaii, where two polls led me to guess an Obama margin of 27 points that was actually 15.8 points too Republican (the real margin was 42.8).
As bad as my results were in the presidential race, my results in the 33 senate races were worse.
There is a systemic leftward shift in that data, which means the actual results were more Democratic than I had predicted.
Here's the same graph zoomed in on the battleground states.
I predicted a 6 point race in Missouri (that dot way at the top of the zoomed in map) that was actually a 15.5 point walloping by Sen. Claire McCaskill. The model only missed on Heidi Heitkamp; I had predicted a 4 point loss. But my average error more than makes up for the fact that it correctly called 32 of 33 races: I had a house effect of 4 points towards Republicans. The only Democrats who won by less than my prediction were Sens. Sherrod Brown, Martin Heinrich and Tom Carper, and I was too kind to Bob Kerrey, who lost to Sen.-elect Deb Fischer by a huge margin. But that's it. Only four races out of 33 were redder than I expected, and 29 were bluer.
No comments:
Post a Comment