Cafe Talk | Talking Points Memo | Median-Weighted Meta-Poll

Median-Weighted Meta-Poll

By John Zapisek - June 25, 2008, 11:17PM

After the 2004 elections, astrophysicists J. Richard Gott and Wes Colley found that the results matched almost exactly a new poll-based indicator they were looking at. In their method, each state was assigned to either Kerry or Bush according to the number of polls each was leading in, irrespective of the size of the lead shown by any of the polls. The only state that differed from their predictor was Hawaii.

Wes Colley now has a web page showing a tracking meta-poll of the Obama-McCain race. You can find it here.

As of today (2008-06-25 Wed), it shows Obama ahead 355 electoral votes to 183 for McCain. States that are shown as flips from their 2004 results are CO, FL, IN, IA, MO, NM, OH, and VA for Obama, and none for McCain.

Of course, the general election is far too distant for this or any poll to be a useful predictor of what will happen in November. So much can happen between now and then. It can, however, serve either campaign as a guide to where to spend resources. By clicking on one of the dates (e.g., 6/25), you can see for each state how many polls favor one candidate or the other.

Their method is described in a 6/6 NYT op-ed piece, and Dr. Gott can be heard discussing it in a Scientific American podcast.

Enjoy! --jzap

Comments (5)

Easily manipulated by adding biased polls on either side.

Posted by Marquis de SeaToShiningSea
June 25, 2008 11:26 PM | Reply | Permalink

Yes, that's possible. The poor choice of a new poll for inclusion would definitely devalue the results.

AIUI, Colley is using only the polls reported on RealPolitics. If that proves problematic, I'm sure he can adjust his selection criteria.

I'd be floored if his interest in manipulating the results exceeded his geek interest in seeing his mathematical method be proven valuable. Astrophysicists are like that :-)

Posted by jzap
June 26, 2008 11:39 AM | Reply | Permalink