Post-election recap
I feel great about how this site performed in this election. The key points are:
- The estimate of 365.7 expected electoral votes was extremely close to the actual outcome of 365.
- The only individual electoral votes I mis-called were MO and NE2. For MO, I only asserted a 76% certainty of Obama winning, so I don't feel too bad that I mis-called it. I had called a 10% chance of Obama winning NE2.
- The final 6.0% popular vote spread was well within my margin of error. I had predicted 6.7% +/- 3.0%.
- Individual state predictions went well, particularly for close states. My average error was 3.61%, which is similar to 538's average error of 3.64%.
- There is a high correlation (+77%) between how well Obama did in each state and how much he exceeded my expectations by. Here are the states where my predictions of the Obama-McCain spread were off by over 5%: AK, AR, DE, HI, LA, MA, MS, NV, ND, OK, VT, WV, WY. In each of these states, my prediction underestimated the amount of the winner's victory. And only NV and ND were real battlegrounds this year. My best idea for an explanation is that these are states where one party has a dominant machine which is more effective at voter turnout; however, that idea can't account for AR, which is dominated by Democrats at all levels except the presidential.
- Because the above effect exists across many states, it would be foolish to use my prediction (or any similar site's prediction) as evidence that vote fraud existed in AK.
- Because I was too busy campaigning to do data input, the last analysis I did on 11/3 did not include data from national polls for the last two weeks. I intend to rerun the process this weekend to include that information, so that I can get a better sense of how the process did.
Read more (maybe)!
posted by Benjamin Schak at 1:27 PM
30 comments
