I was just going through the polling for Gubernatorial and Senatorial races to update my predictions, and looking at Rasmussen really worries me. In a lot of states, they are the odd man out in predicting Democratic victories. In fact, in most instances, his one-day polls are finding results that are distressingly conflicted with other polls, mostly in the Dem's favour.
While Rasmussen does have a history of slight Republican lean, I've never seen Scotty move around this much. This one-day polling is really bothering me and, despite good 2004 results, I'm having trouble trusting Rasmussen - a shame, since they are so prolific.
Any comments on this?
I reposted some old Vorlon commentary about his discussion with Rasmussen per 2004 state and national polls. Hope you noticed it.
I'm basically going to repost some of my own research in combination with his knowledge to decipher a conclusion.
The big difference between 2000 and 2004 is that he stabilized or weighted his national turnout model towards expected turnout numbers (sort of like how Zogby does it, but without as many useless weights as Zogby) and nailed it.
In 2000, he admitted that his turnout model underestimated Dem turnout. He wasn't the only one, as Dem turnout was underestimated in that race till the end. Same thing happened with the Battleground poll, which had been the most accurate in the 1992/1996 election period, got the turnout model wrong in 2000, but refigured and got it right in 2004.
But anyway, this has nothing to do with his state polls, which operate much differently.
Remember, Rasmussen conducts his polls very similar to SUSA, using "cluster samples" to generate his polling sample. I've tended to find that this method can jump around like crazy early in the campaign, but is usually fairly accurate in the end. Similarly, I don't trust his one-day polls right now, unless they show the same thing consistently. Same can be said for SUSA. I also think that the Rasmussen method works better when conducted as a tracking poll, rather than one one-day poll per month.
Mason-Dixon is much more accurate over the whole of the campaign because their method is simply much more solid and reliable.
Actually in 2000, Rasmussen's state polls performed very well as I recall, getting 15 out of 17 states correct and most within MOE. This was, of course, overshadowed by the national problems. He did well in 2004 also, but we all know that.
So, in the end, that's what I look at. My predictions are giving some credence to polls right now, but I'm also keeping in mind the weakness of polls this early in the campaign and concerns with regards to wild jumps. My predictions haven't changed as wildly either because of this.
