WI: Other Source: Obama up 15 in WI

[Gern Blandsten]

Ah, the DeSart-Holbrook model

Still not good.

Not good?  It doesn't look much different than everyone else's.     

The only thing in their prediction that might be suspect is wv, and possibly mo.  Everything else looks pretty solid.

[Gern Blandsten]

Ah, the DeSart-Holbrook model

Still not good.

Not good?  It doesn't look much different than everyone else's.     

The only thing in their prediction that might be suspect is wv, and possibly mo.  Everything else looks pretty solid.

Their model is flawed, whether or not it gets good results.

Well, that's a pretty empty statement.  By your logic, even if it generates perfect predictions it's still flawed? 

[Gern Blandsten]

Well, that's a pretty empty statement.  By your logic, even if it generates perfect predictions it's still flawed? 

Yes; see TheGlobalizer's analogy.  If the foundation of a model doesn't make sense and can't be reasonably explained, getting lucky does not make it a good model.

And every model is flawed, this one moreso than others.  It is, IIRC, primarily based on unchecked but time-weighted polling information (hence WV), which is cool but not an especially good model.

So, by your standard, how many times does a model have to be accurate (and how accurate must it be) to pass the "randomizer test"? 

It seems to me that the logic underlying the model makes perfect sense.  Candidates tend to win states in which they hold the lead in September, while controlling for the historical leaning of the state.  I'm not sure if they weight the poll data, I don't think so.   Not sure what you meanby "unchecked."

I saw them present it at a conference a couple years ago. It was a real simple model, and I think they were real close to the actual result in 2004.  R-Squares over .9