Thursday, October 28, 2004

31% Chance of Kerry Presidency
from the mind of  Daredemo.

That's my prediction. How do I get this? Well, I decided while waiting for some code to run this evening to cook up a simple electoral college simulator. I went over to the RealClearPolitics site and recorded all the state by state polls they had collected in the past two weeks since the last debate. In states where multiple polls existed I combined the percentages and errors to give me a sort of combined measurement of how the vote would go in that state. I then told my computer to use that information to randomly generate 9999 fake elections (be careful how you say that), and calculate the resulting electoral college totals. It took roughly a minute and a half for it to perform all of them, as there were no court cases or recounts in my simulation. My results are summarized in the plot below:

This is the distribution of the number of electoral college votes President Bush would get assuming the errors in the individual polls were Gaussian (probably more or less reasonable), and assuming that any fluctuations between states were not correlated (hard to say, maybe thats not so reasonable, but its the easiest thing to do).

The red colored guys on the right are those that went for Bush, blue on the left are those that went for Kerry. Roughly 67.5% of the 9999 simulated elections I ran ended up as Bush wins, 31% as Kerry wins. That leaves 1.5% -- which are the guys that ended up with both candidate getting 269 votes -- an electoral tie! (the thin yellow stripe in between the red and blue)

So to be more or less official then, lets say my prediction for the number of electoral votes Bush will get is (drum roll): 281 +/- 24. I also predict a 1.5% chance of an electoral tie. In which case I believe the Republican led House of Representatives gets to choose the President...

Before people start taking what I've done here too seriously, let me point out a series of caveats:

  • I used all state poll data posted on the RCP site dated from 10/14 - 10/28 and assumed all were accurate, no matter the source of the poll. All polls were assumed equivalent and independent.

  • I treat Maine and Nebraska as all or nothing states for simplicity, since they both are on opposite sides and roughly have the same number of votes -- I also assume Colorado stays as an all or nothing state.

  • Nader and undecideds were dropped in each poll -- I.e. if a poll was taken of 1000 people, and it was listed that 50% said they'd vote for Bush, and 48% said Kerry, I then would use 500 "Bush" votes, and 480 "Kerry" votes to calculate my vote ratio and errors. The remaning 20 would be dropped. I think this is fine, as we only want to know who wins each state. A warning though -- I've read that historically undecideds tend to side with the challenger rather than the incumbent (the theory being if they were completely happy with Bush they wouldn't be undecided). This would push my distribution to the left slightly.

  • In some states the latest polls were from September or early October. All of those cases were states with low numbers of electoral votes, and high ratios toward either candidate (usually Bush). I took them at face value.

  • No attempt was made to try to use any information on trends in the data when multiple polls were available, though errors in these polls (typically around 4%) really are probabably too large to even try.

That all said, I believe looking at the numbers this way has much more meaning than simple, across the board national polls like we see all the time. Dunno about you guys, but this 31% (or possibly a bit higher) chance of a Kerry presidency has me rather scared. That's a bit high for my taste... Here's hoping the weekend talk shows spend their time examining the current terror threats and get some people to come to their senses about what's at stake here...


Blogger Evan said...

This is great! Thanks! And... people say that conservatives have no grasp of science. I'd say the reality is that conservatives have a firm grasp of science, but we refuse to be a victim of 'science'... (notice the scare quotes.)

9:59 AM  

Post a Comment

<< Home