Tuesday, September 30, 2008

Democrats pulling ahead in Electoral Vote simulations based on prediction markets

This week's Electoral Vote Simulation is starting to show a wide lead in the electoral votes for the Democrats. I chalk this up to the Intrade prediction market data for each state beginning to reflect recent polling results which favor Obama and that are lower for McCain. This data should have taken into account the recent financial crisis in the news, McCain's strange campaign and debate off-again on-again gambit, and the debate itself. I think that these results offer some proof, along with the polling data, that Obama "won" the debate and that the economic issues are favoring him.

The median electoral vote result from 1000 simulations is 303 votes for the Democrats, while 85% of the simulations show the Democrats with more than the 270 electoral votes needed to win. This is the highest margins since I started running these simulations at the beginning of September. These results are also reflected in the strong vs weak Democratic or Republican states as seen in the table below.

The states that are weakly Democratic are Nevada, New Hampshire, Ohio, Virginia with new addition Florida pulled from the Republican side, the one that is weakly Republican is North Carolina. This shows a strengthening of Obama in formerly weak Republican states, and a weakening of McCain in formerly strong Republican states. I am sure there will be some regression to the middle before election day, but this is bad news for the Republicans no matter how you spin it.

For comparison I have included links below to much more professional interpreters of election and polling results. Other sites with electoral vote simulations of the election are:

FiveThirtyEight.com which has Obama at 329.3 electoral votes to McCain's 208.7 and the Democrats winning in 83% of his 10,000 simulations. This site uses a simulation based on the polling data and uses models to weight polls differently and to use regional information when individual state polls are unavailable or outdated.

Electoral-vote.com has 286 electoral votes for Obama, and 225 for McCain with 27 in a tie. This site also uses state by state polling data.

RealClearPolitics.com has Obama with 249 electoral votes and McCain with 163 with 126 in the toss-up category. They also use polling data.

270towin.com uses the last 1000 simulations in their visualizer to build up a results that has Obama with 298 electoral votes and McCain with 240. They have the Democrats winning in 85.6% of their simulations. They also use probabilities of each state winning based on polling data with some clipping of the data.

The Truthisall 2008 electoral model has Obama winning almost all of their 5000 simulations. They seem to be real concerned about election fraud and have a hard to understand election model with assignments for undecided voters.

I have also mentioned getting my data from the Intrade Prediction markets, and Yahoo's Political Dashboard.

I invite you to form your own opinions about the various electoral vote models across the web. I run my model for fun and I sue the Intrade prediction market data because it is very easy to turn into a simulation. Some have issues with this data. I don't have the time or the energy to review the state by state polling data to create probabilities for simulation, and I think that FiveThirtyEight.com has done a very good job of doing that. I just think that the model doesn't seem to be transparent enough for me to duplicate it as you could do with mine so it is hard to check the results or evaluate the assumptions. 270towin.com seems to have a similar good approach, but still lacks transparency in the model, and have their own opinion about handling the data. The other models fall further away from these good examples.

My idea is to gather all of the voters on November 4th, count their results, assign electoral votes according to the rules in each state (winner take all, except for Maine and Nebraska) and see who wins. That results will have the least error (though there will be some).

No comments: