Tuesday, September 16, 2008

Presidential electoral vote simulation predicts Republican Victory

The Palin Republican Convention bounce seems to be having a continuing effect on the presidential electoral vote simulation based on the Intrade Prediction market data of September 16th. As described earlier, I use the Intrade data to generate a probability for each state of tghe Democrats winning that states electoral votes. I generate 1000 Monte Carlo simulations based on these probabilities.

This week the chart shows that the the Democrats win with more than 270 electoral votes only 46% of the time. This is the first time that the simulations predict the Republicans winning more than half of the time. I think that we could see this coming as the prediction markets and polls continue to digest Sarah Palin's popularity and McCain's bounce. FiveThirtyEight.com shows similar results based on their excellent simulations based on state polls, with only 44.6% of simulations with an Obama victory.

As a prisoner of Excel as my math, simulation and graphing program, I am jealous of Yahoo's new election tracker dashboard that uses either state polls or the Intrade prediction market data to show the electoral vote results. It made me notice something interesting about the Intrade prediction market data. If I just use that data, as Yahoo does, to predict whether a state will go Democratic or Republican, the results is that the Democrats win with 273 electoral votes. But the prediction markets have been shown to represent the collected estimated probability that the players expect for a particular outcome. So while Delaware options are at 95cents (on the chance of $1.00 if the option pays) for the Democrats which makes that a pretty sure bet, Colorado is 55cents Obama vs 47.6cents for McCain which makes that close to even odds. That's why I like my simulation approach better.

The table below (click it for larger and more readable) compiles the simple sum of the prediction market data and then breaks out strong states, those with 60% chance of one of the sides, vs weak states, those with less than a 60% chance for one of the sides. It is there where you can see the closeness of the race.


Remember that in the simulation even Delaware could go for the Republicans 5% of the time. Highly unlikely but possible, especially with the potential error of the data we are using. A state like Colorado could go either way. The progression over two weeks since Sept 2nd doesn't seem to put more electoral votes in play in the weak states (51 to 56), but it does seem to move some weakly Democratic ones to the weakly Republican column. This week the states that are weakly Democratic are Colorado, New Hampshire and New Mexico, while the weakly republican states are Nevada, Ohio and Virgina. On Sept 2nd the states that are weakly Democratic were Colorado, New Hampshire and Ohio, while the weakly Republican states are Nevada, and Virgina. As the pundits have said, Ohio and Virgina are going to be important, and even some Western states are going to have a role to play

1 comment:

TruthIsAll said...

The Election Model uses a 5000-election trial Monte Carlo simulation. The model projects that if a fraud-free election is held today, Obama would win 323–215 Electoral votes with 51.1% of the two-party vote. The EV win probability is a simple calculation: Obama won 4926 of 5000 simulated election trials; his win probability is therefore 98.5% (4926/5000). It’s a snapshot which changes slightly every day.

The model indicates that for the same 303-235 EV split, Obama’s EV win probability is 92% (assuming he wins just 50% of the undecided vote). Since the probability calculations in both models are based on the latest state polls, there is obviously a difference in methodology between the models.

The Election Model base case scenario assumes that Obama will win 60% of the undecided vote. And this is conservative, as he is presumed to be the challenger (McSame is running for the third Bush term).

View the Election Model Electoral Vote Simulation Frequency chart. Note that 4926 (98.5%) of the 5000 simulated election trials are over 270 for Obama. Compare this result to the equivalent fivethirtyeight.com chart in which 28.5% of the trials which McCain won are in red, while the 71.5% Obama won are in blue. The chart should be 98.5% blue.

Obama also leads the National projection model (based on the average of the latest 5 national polls) with 52.4% of the 2-party vote. Note that the national polls lead the state polls, so that we can expect a rise in Obama’s expected EV and win probability. The national model also assumes that he will win 60% of the undecided vote. The probability that he will win the popular vote is over 98%.

http://www.democraticunderground.com/discuss/duboard.php?az=view_all&address=203x508448