Friday, October 29, 2004

Probabilistic Election Model

Q and O provides a link to a probabilistic model of the election, done by Stanford. I like this model better than the models based on polling data (and not just because it's predicting Bush will win, either, since all the ones I've seen also predict a Bush win.) No, I like it because it's based on probabilities. See, even in a "statistical dead heat," there is always some probability that your guy will win, which can vary quite a bit. For example, if Bush leads Kerry 50 to 48 with a 3% margin of error, you have to consider not only the scenario where Kerry wins 51 to 47, but also where Bush wins 53-45, etc. etc. It's not always a 50% chance that your guy will win just because the poll has it statistically tied. Not only that, but the very idea of the 3% confidence interval is that 95% of the time, the actual numbers fall within that 3%; that leaves the 5% chance that they don't. This model takes all of that into account.