"The world is not a lodging-house at Brighton, which we are to leave because it is miserable. It is the fortress of our family, with the flag flying on the turret, and the more miserable it is the less we should leave it."
-G. K. Chesterton, Orthodoxy
This here blog is a glimpse or two or three at the condition of the 'fortress of our family' through the eyes Timothy Goddard, a Christian writer with an unhealthy interest in politics living in the Puget Sound area.
Projections as predictors (and redundancy as repetitiveness)
This morning in my analysis of three electoral vote projection sites, I cast very mild aspersions on ElectionProjection.com, and specifically, the formula used to make the projection. I had previously held the site in the highest esteem, and only questioned the formula when I noted that it was moving states like Oregon and New Mexico into the Bush column, despite the fact that no polls had shown that.
Then, this afternoon, MSNBC released their Blue State battleground polls (they had released Red State battleground polls the day before), giving Bush a 4 pt. lead in each of those states. I hastily posted an update, but was curious. The poll didn't appear to have come out until about 7 pm EST, but Scott at ElectionProjection had posted his update yesterday. Had I missed the poll, had he seen it early, or did his projection actually project something true?
I emailed Scott about this, and this is the reply I recieved:
I didn't see the polls released for Oregon and New Mexico before you did. I think the predictive nature of my formula, if there is one, is that it infers from overall job approval and national head-to-head polls where each state should be. In other words, if Bush lost a state by 5% in 2000, but leads the national polls by 5% this year, the formula would project that state to be about even. So far, it has seemed to flow well with state polling data. The exception is when extraordinary circumstances influence a particular state in a way that causes a fundamental shift in the sentiments of the electorate. An example this year is New Jersey. Because of its proximity to the attacks on New York City and McGreevey's problems, Jersey is polling much better for Bush than my formula projects.
I think another positive attribute of my formula is that it doesn't fluctuate wildly as different polls are released. A lot of diverse information is used, so one or two outliers don't have drastic effects on the numbers.
Obviously, he's onto something here, and his site has quickly become again my favorite projection site after a very brief hiatus.
As Scott notes, his formula suggests where a state "should" be, but unfortunately, has no frame of reference for that outside of common sense (which appears to be doing pretty well for him right now). I hope that, after the election, he will fine tune his formula for 2008--though, without an incumbent, more drastic revisions may be needed.
Posted by Timothy8:30 PM
Agree, disagree, have more information on the topic? Please, feel free to leave a comment. No profanity!