As I'm sure a lot of readers already know, Georgia has an unusual provision for their senate elections. If neither canidate receives over 50% of the vote, there must be a runoff election.
Since our model only uses two-way vote share as an input, this breaks our model without further modification. Luckily, substituting it with a three-way vote share in the Georgia race(while still stripping out undecideds) leaves a well behaved stochastic process that can be plugged into our Bayesian filter without too much difficulty.
We've talked about this earlier, but we've incorporated the new polls into our forecast.
+/- series show 95% confidence intervals of the polling aggregate from our filter
Same graph as before, but zoomed.
There are three realistic outcomes in this race(The chance of Buckley winning is essentially zero)
1) Martin receives over 50% of the vote tomorrow, winning the Senate Seat for the Democrats
2) Chambliss receives over 50% of the vote tomorrow, winning the Senate Seat for the Republicans
3) Nobody receives over 50% of the vote tomorrow, and a runoff is held in December
According to my Bayesian filter, the probabilities of those outcomes are roughly 3%, 43%, and 54% respectively. In pie-graph form:
So, this blog forecasts that neither side will reach 50%, and that the election will lead to a run-off. We're incorporating this into our Senate Projection.