SIMULATING THE PRESIDENTIAL RACE
So, I have lots of data. I have means and margins of error. I have state by state numbers in the Composite analysis. So, I have enough to construct probabilities and run simulations of the presidential election.

Over on DU, I found a fantastic effort by DU member TruthIsAll. I have a few problems with his state by state probabilities because of several assumptions that provide the basis for his simulations. One assumption is that undecideds will break heavily for Kerry, 70-30. I previously discussed why that wasn't reasonable. I'd like to see him run his model assuming undecideds go 55-45 for Kerry. He also determines the likely Kerry Vote for 2004 by looking to current state polls and the vote for 2000, 1996 and 1992, picking up 3 Democrat popular vote wins, and he doesn't normalize each state to the national popular vote, which might give a different picture. He should also add in 1988 results, at least, to get two Republican victories and 2 Democratic victories. In fact, if you study the voting trends per state, using data from 1988 improves the accuracy of predictions of 2000.

Anyway, TruthIsAll sets the current probability of a Kerry victory at 97.4%, based on his assumptions. But a look at the constructed probabilities by state don't seem reasonable. Is it really an 83.5% likelihood that Kerry will win Arizona, where recent polling shows he's usually behind and Bush has recently been polled with double digit leads there? Sure, a recent poll showed Kerry up 1, but an 83.5% probability? TruthIsAll needs to audit his data applying a little reasoned analysis and not just number crunching.

I started working up a similuation based on my Composite Poll results. Currently, Bush has a probability of winning Arizona under my system of 62.5%. Which sounds much more reasonable to me.

As a tease, I'll tell you that my current analysis of 10,000 election simulations says that Kerry's probability of winning the election is 56.0%, with an average of 272 electoral votes and a maximum of 371 electoral votes. I'm still auditing the methodology to make sure the probabilities pass the smell test and hope to get all the data up in the next week.