Rasmussen released a new poll on the Virginia Governor's race showing McDonnell with an 8-point lead (49-41). The poll is mostly useless, however...I'm not a Rasmussen subscriber (I'm not made of money), and I don't have access to crosstabs, so I can't evaluate the poll in depth.
What I can say is this: Rasmussen has been more than a little funky lately. Something seems to be causing a systemic skew to the GOP in his polling. This shows up best in Obama job approval ratings, where he has continually undershot the rest of the pollsters by large amounts, regularly yielding a net approval rating 10-15 points under almost all of the other pollsters. This also shows up in his PA Senate race polling (Toomey up by double digits is a bit hard to believe; I can believe a lead, but not one that's so large) and his Generic Congressional Ballot (which he does more of than anyone else).
With this in mind, all of Rasmussen's polls strike me as dodgy, not just this one. Something is fishy here, and it's been puzzling me for a while.
I've got a few guesses:
- The first is that he has a different sampling/weighting model. Not wrong, just different. I can't say it's wrong because I haven't looked at it, and it has worked in the past. So the model may have something to do with it.
- Second, we have sampling problems. Pollsters are having more and more trouble getting people to answer the phone. I think this was responsible for the SUSA and PPP age distribution problems in the last VA polls they did. Younger voters tend to break a bit more Democratic in most polls you see, so if he's undersampling and not adjusting for that (or considering biases among who he's getting in that crowd), then that could be partly to blame.
- Third, there's 1 and 2 compounding eachother. A model with a slight bias in it can lead to a real problem when something else exaggerates that bias (i.e. his "ideal sample" is weighted wrong, but then sampling biases are making that problem worse).
- He's seeing what he wants to see. I don't want to accuse any pollsters of taking a sample at face value because they like what they see, but there's a chance that unless and until his model is shown to be wrong, Rasmussen isn't going to go "under the hood" and tweak his sample. Also, it may be commercially viable for him to put his stuff out there...he's certainly got good clients with FOX if he puts his stuff out there.
I'm not going to comment on where I think Rasmussen is compared to the actual state of the race. I think the issues in his recent polling work make it useless without crosstabs to evaluate. He's clearly got a model, but without some access to the data which underlies his results I can't do anything with his polls except to issue a caution against putting too much faith in anything Rasmussen puts out. With the other main pollsters, I can at least get an idea of whether the poll is rogue or the sample is funky; with Rasmussen, I get the feeling that the pollster has gone rogue.
Unless and until I can sort out what's up with Rasmussen, I'm going to largely set anything he does aside. If anyone knows what's up here so I can actually evaluate his polling, I'd greatly appreciate it.
Edit, 3:10 PM: A good point was made in a comment that made me think. Therefore, I'd like to tack on here: Rasmussen's polls are likely invalid, but they are constantly so. Thus one might be able to draw a trend out of what he says, but you'd have to deduce the exact balance of the race indirectly. As such, I think the term for Rasmussen's polls is that they are reliably invalid. I still plan to set them aside because of the validity issues, but...hey, at least if he's bumbled his model, he's bumbled it in a very consistent way.