Hominid Views

After spending 13 months scrupulously collecting state head-to-head polls every single day, including weeks while I was on the road), and analyzing them almost every day new polls were released, the election is now over. My first reaction is “freakin’ finally!” But, intellectual curiosity wins out over…um… exhaustion, and I am curious to see how the predictions worked out.

One problem is that there are numerous types of “predictions.” One can predict the outcome of each state and one can predict the electoral vote total. The two predictions are not really the same.

Here is why. Suppose you have a coin that turns up heads 60% of the time. Clearly, for a single flip of the coin, your best prediction would be for a head to turn up on a single flip.

Now, suppose you want to predict the result of an experiment where you did 51 independent coin tosses. Your best prediction for each individual flip (before the fact, of course) would still be a head. But, you would certainly not predict 51 heads as the outcome of the total experiment. That would be silly. Rather, you would estimate an expectation (as a statistician would put it—or an average or a mean in the vernacular).

For this particular experiment, the expectation for 51 tosses with a probability of heads at 0.60 would be 30.6 heads and 19.4 20.4 tails. That is not the same as the sum of the individual predictions.

So, in this same way, there is a difference between predicting the state-by-state wins and predicting the electoral vote total. The former is about the most likely outcome for each state, and the latter is about the collective results.

How’d I do then? Unfortunately, Missouri and North Carolina are very close and headed for recounts. But it sure looks like Missouri will go for Sen. John McCain and North Carolina will go for President Elect Barack Obama. Let’s suppose this is the outcome. That leaves the Nebraska NE-02 congressional district that is really close as of this writing, but favoring McCain—let’s pretend McCain takes it. (Nebraska is not a winner-take-all state, so the state’s votes can split among the candidates.)

The electoral vote total should be 364 for Obama to 174 for McCain.

Here is my final map:

The only individual state I predicted wrong was Indiana. The six Indiana polls leading up to the election showed somewhat mixed results, but with two late Zogby polls giving McCain a +5% margin, it is hard to argue that the evidence supported an Obama victory. It didn’t. Oh well…lots of us got this state wrong.

By the way, I am particularly proud that my state-by-state predictions were better than Karl Rove’s. The difference? Only North Carolina. I got it right…Turd Blossom got it wrong.

What about the electoral vote? Here were my predictions:

Some electoral prognosticators simply tallied the electoral votes in the projected winning states. Karl Rove did this and came up with a prediction of 338 for Obama and 200 for McCain.

That was not my strategy. Rather, I computed an expectation over “all” possible outcomes. Literally, I took the average of the electoral vote outcomes from 100,000 simulated elections. As a result…I predicted 364 to 174. As it happens, that result only occurred in 4.3% of the simulated elections in my final projection run, but it was, statistically, the expected outcome.

So…I’m not really the self-promoting kind, but it looks like I nailed it! At least I did if Montana, North Carolina and NE-02 continue to cooperate with me.

Stop back soon, when I will compare my results, methods, and scientific modeling philosophy to that “other prediction site,” fivethirtyeight—you know…that site that has the byline “Electoral Predictions Done Right” and predicted 348.6 for Obama to 189.4 for McCain….