This blog is not about politics. Today's post will briefly look at predictions for today's elections across the US, but only because there is one site that takes a scientific look at the polling data and uses it to make projections. That site is FiveThirtyEight.
The details of the methodology used are somewhat unclear (certainly for porprietary reasons) but the blogger, Nate Silver, a baseball statistician, uses a Monte Carlo method to run simulations using published polling data. This election cycle is actually his second time out - the first time was 2 years ago and covered just the presidential election, and in that case, his model was pretty accurate in predicting the electoral college results (hence the title of his blog). This time, he's looking at the governor races, senate race and the representatives too. I think this will be much more challenging as the number of polls for many of these races are quite a bit smaller.
Regardless, the outcome of the 100,000 simulations is a distribution with an average and a range of what we can expect, all of which is more meaningful to us scientifically inclined individuals than what is reported by the talking heads. If you've read his stuff in the past, you know he personally has a leftward bend in his politics, but to me, he really tries to be as accurate as possible with these predictions. He is very up front that for 10 races that he is 90% certain of the winner, he had better be wrong (or near wrong - that's just the way statistics works) in one of them, or else his model is wrong.
I will be quite curious to see tomorrow how well the predictions come off. As of this moment in time, he has the Democrats as 93% favorites to control the Senate, but only a 16% of controlling the House. He has odds for all the individual races as well, so check out your local ones for yourself.