Tuesday, May 20, 2008


Fivethirtyeight.com gets crushed! tonight

Regression analysis is not all that it turns out. For those who follow these posts, I previously praised fivethirtyeight.com with its remarkable calls in Indiana and North Carolina....

Well now comes the boom. http://www.fivethirtyeight.com/2008/05/kentucky-projection-clinton-by-19.html

Clinton won by 35 points tonight... 35. Not the 19 predicted by the mystery man/woman at fivethirtyeight.com. Not many people will notice this large gap because well 35 vs. 19 is much different in terms of perception than 19 vs. 3.

I'll go through why he got it wrong at a later date... but I think poblano's own analysis warrants merit:

"At least in terms of surface-level demographics, West Virginia looks something like Kentucky less Lexington and Louisville. But Obama actually backtracked about 10 points in those West Virginia type of regions from last week."

Surveyusa, who bit the big one two weeks ago, called for a 31 point Clinton victory.

But the big prize of the night goes to American Research Group... who called for a 36 point Clinton victory... and nailed her percentage at 65. For a polling group, most followers dismiss.... Dick Bennett has actually called the races closer than Poblano the past two weeks... In WV and Kentucky.....

Not bad.

I don't make this posts to dismiss Poblano... and I certainly take his/her word over Dick Bennett... but I do want to say, that my faith in polling has been restored (if it was ever missing)... and although regression can be a powerful tool.... electorates are sometimes just flat out unpredictable.

Comments: Post a Comment

Subscribe to Post Comments [Atom]

<< Home

This page is powered by Blogger. Isn't yours?

Subscribe to Posts [Atom]