Electoral-Vote.com Prediction

[davidjhorrell]

Why does this:

http://www.electoral-vote.com/fin/oct10p.html

Get predicted, given our current lead? (http://www.electoral-vote.com/) I don't understand how this thing works. The system of using most recent polls seems pretty junky to me anyway, but if someone could explain the map prediction, I'd be grateful.

[spotbird]

1. That's not what's coming up for me.

http://www.electoral-vote.com/

Try this one.
http://www.race2004.net/

[davidjhorrell]

2. Much better!

:)

[Arancaytar]

4. The other one is the projected final map.

It appears to be calculated from certain trends in the polls, which I don't understand. After the last week, the trend should be much more favorable, given the boost Kerry got. Still, the projection is far worse than the actual map.

What worries me is that the boost is wearing off again. A few days ago, Kerry has 280:235. Yesterday, it was 280:245. Now it's 270:248. I hope the third debate will kick Bush's ass so far that the lead goes back to what it was before.

[DuaneBidoux]

3. I agree this is weird

They must have been projecting (from sometime in the past in September) what they thought it would look like in October. I will tell you one thing, based on the Zogby and some other state polls I've read I'm much more inclined to believe the one on the main page.

[Ediacara]

5. It's a least squares regression

The idea is to get the equation of a line where the vertical distance between the line and points is minimized.

This is kind of neat, but also kind of useless for three reasons:

1) It's terribly succeptable to outliers, as a bogus poll (like say, the Badger poll in Wisconsin that showed Bush ahead by 20) will force the line to fit. A major problem with electoral-vote is that he refuses to weed out clearly bullshit polls.

2) It's linear and polling trends aren't necisarily linear.

3) If there is a late Kerry surge it'll be underrepresented by the least squares regression.

[unblock]

6. determining a trend is VERY touchy, given the sensitivity

make a different assumption in how to weight the different historical polls and your "trend" can tilt quite a bit one way or the other.

given just how close many states are, a very slightly different formula can skew the results quite a bit.

i wouldn't pay attention to "trend" projections until the last week or two before the election, and even then, the above caveat holds.

bottom line, i don't think "trending" the polls is very useful analysis.