sock puppet wrote:
Interesting that the PollyGraph linked shows the predictability of Obama's 2012 re-election having been 53% when it began in April, oscillating and reaching a high in early June just above 54%, and having taken a fairly steady downward trend since then to 50.2% today.
What that graph shows is a projection of the Republican/Democrat share of the two-party vote. Theoretically a 3rd party could alter odds of victory despite the graph itself being accurate. The closer and closer you get to election day, the more accurate that graph gets (to the point of frightening accuracy by the end.)
There are a variety of predictive techniques that this system melds. But, there's actually a pattern to how effective each technique is depending on where you are in an election cycle. They're all very accurate within a certain margin of error. Right now, index models are the most predictive. In a few months, econometric models will overtake them in that category (within a few points). That will be closely followed by market betting and finally sophisticated extrapolation from polling (what Nate Silver does at 538). By hedging these models in the right proportion against one another, you get the best political forecasting.
Right now, the model predicts a very narrow Obama victory of the two-party vote.