Thursday August 23, 2012
I’ve added an Electoral College counter to the code running the poll averaging I’m doing for HuffPost.
This is the usual, posterior/predictive guff: (1) we sample many times from the posterior density for Obama and Romney vote shares; (2) for a given sample, is Obama > Romney in state s?; if so, give the EVs for state s to Obama, otherwise to Romney; (3) sum over states; (4) repeat many times over the many samples from the posterior density. Make a pretty histogram when you’re done. Count the number of times you see Obama at 270 EVs or better. The proportion of times this happens is the posterior probability that Obama wins the Electoral College (at least given your model and its many assumptions, uncertainty over its parameters, etc etc). Note no special treatment of Maine, Nebraska.
Here’s what I get, or got, middle of the day, Thursday August 23; Obama narrowly ahead, with a .684 probability of carrying the Electoral College, but a better shot than that currently on offer at InTrade (.572). Note that the InTrade number is more properly a forecast than what comes out of my “pooling, adjusting, smoothing, x-state leveraging” of the polls (and relying on historical election results, etc); punters get paid on what happens in November, not on what poll respondents say now.