Polimetrics

Projection 10/19 4PM PDT.....

Kerry 59,001,000 50.4% 291 EV
Bush 56,470,000 48.2% 247 EV
Other 1,689,000 1.4%

After months of waiting, I have published the first battleground projection - better late than never! The poll shows a near dead heat (big suprise) - with Kerry up by 0.6% in Florida, the current decisive state.

What makes this projection different than the hundreds of projections already out there, many of which I have linked?

1. The projection uses a partisan index - a best estimate correlation among the states based on past elections and demographic projections by the US Census Bureau. For example, midwestern states such as Iowa continue to trend towards the right while states like Arizona have trended left due to internal migration. Using multiple regression on variables such as ethnicity trends, population growth, regional factors, etc, I determine a partisan index - what many call blueness and redness. Final results have been adjusted for the candidate's home region based on past elections since 1980.

2. The projection takes into account national polls, for adjusting the partisan index to current status and state polls to adjust the partisan index within its determimed 95% confidence band (about 3-5% based on past performance.) In other words, because the methodology for state polling is not consistant, I use a Baysian approach in adjusting all state results by the partisan index. For example, in this run the most extreme adjustment was South Dakota which has an adjusted partisan index of about 28% but has reported a state poll with a 10% difference. The actual result of +18.6 is a more reasonable estimate than the reported poll. Note, these adjustments are automatic based on model parameters - not based on any subjective analysis.

3. Undecided voters are assumed to break in accordance with past elections when one of the candidates is an incumbent. Third party voters are determined based on an expected national vote of about 1.5% adjusted by each state's third party relative voting trends since 1980.

4. Using US census data, I have also projected total votes for each state to determine a predicted national share.

Why I am I doing this?

I am a Math/Statistics professor who has always been interested in the study of political statistics. This is my latest attempt at making an empirical tool to predict and study presidential elections. I hope to do for politics what Bill James has done for baseball - to educate people, expose misconceptions and ignorance, cut through the rhetoric, and to examine politics using purely statistical methods. Most of all, I want to have fun with this - politics is the most entertaining game in town.

For the record, I am not unbiased - who is today? I have already cast my ballot for the challenger in this election. My rhetoric, (probably like yours), is posted all over the Internet at places like Daily Kos. I will keep this place rhetoric free - notice there are many good data links from "the other side" - like "Election Projection", one of my favorites.

When it comes creating political numeracy, I hope we are all on the same side.

Mo