In the R2000 daily tracking poll DemFromCT had the following observation today about the Rasmussen poll:
Rasmussen continues to favor McCain by a few points relative to the others (similar Obama numbers but the McCain numbers are 4 points better than the mean of the other polls.)
DemFromCT then muses:
Rasmussen has an excellent track record, but you wonder about house effect, the tendency for the poll to lean toward
McCain.
If you are wondering why Rasmussen has the race tied today, while every other daily tracking poll has Obama with a significant lead, follow me below the fold...
First, about that excellent record thing. Rasmussen did have some pretty impressive results in the 2004 race, especially in the swing states like FL, OH, and MI, as well as their national vote projection. On the other hand, they stunk up the place in 2000. The National Council on Public Polls had this to say about Rasmussen's 2000 Presidential Poll Performance:
Two other organizations used methods that previously had not been used. Harris Interactive conducted its polls on the Internet among a panel of e-mail users and forecast a tie. Rasmussen's Portrait of America poll was off by 4.5 percentage points on each of the top two candidates. Rasmussen had its interviews conducted by a computer playing a recorded voice with no live interviewer intervening.
That 4.5% percentage for each candidate translates into a 9% error. Oops.
But now to the current race, and why, at least in my humble opinion, Rasmussen shows different results compared to the other daily trackers. It all comes down to party weighting. Rasmussen is seriously tweaking the numbers to favor McCain. Up until September. Rasmussen adjusted their weighted numbers monthly for Democrats, Republicans, and Independents based on their phone calls for the previous six weeks of polling. Now they are adjusting every week, starting in the middle of September. I think those adjustments are bogus, based on their own reporting.
Here are the monthly weight adjustments from April thru August:
August R-33.2% D-38.9% I-28.0%
July R-31.6% D-39.2% I-29.2%
June R-31.5% D-41.0% I-27.5%
May R-31.6% D-41.7% I-26.6%
April R-31.4% D-41.4% I-27.2%
And the current week:
R-33.6% D-38.7% I-27.7%
See a pattern here? In April the difference in weighting between Republicnas and Democrats was 10%. In May, it was 10.1%. Since then the weighting has tightened to the current weekly disparity of 5.1%. But in looking over Rasmussen's site today, I found their daily tracking numbers for June. You can find it here.
The June tracking numbers indicated the following averages for the month:
R-33.48% D-42.84% Net difference-9.37%
Compare this to the actual weighting for the month of July:
R-31.6% D-39.2% Net difference-7.6%
Thats a downward adjustment of 1.9% for Republicans, compared to a downward adjustment of 3.6% for Democrats.
So tell me Scotty, if your own data shows a 9.37% difference (and remember May was a 10.1%), then why did you adjust to a 7.6& difference? Rasmussen has continued to move their weights closer since July. The one exception was the first weekly adjustment in September, but it was so small that it would make no statistical difference.
The Research 2000 poll now posted here on DK uses a R-26, D-35, I-30 weighting. That's a 9% disparity between Republican and Democrat ID. That seems to be about right, and is validated by the generic congressional R-D disparity, which is also 9% (other polls asking the generic congressional question show similar results).
There is another factor that may explain the Rasmussen disparity with other polls, in September they switched from a registered voter to a likely voter model. And while other pollsters, such as the DK Research 2000 poll and Quinnipiac also use LV models, one has to wonder what effect switching horses models in mid-stream might have.
I am not implying that Rasmussen is in the tank for McCain. But it is a Republican polling outfit; that much is readily apparent if you read their headlines. I am suggesting that Rasmussen's current weighting is highly suspect, and is resulting in numbers showing a tighter race than we really have at the moment.