Everybody seems to assume that the results that Rasmussen posts on successive days are based on overlapping data sets,
i.e. that they share two-thirds of their data in common. This is very hard to square with his published results. If one makes this assumption, one gets absurd numbers. Could it not be the case that Rasmussen polls 3,000 people each night, divided into three equal groups: those who will be part of the next days results, those who will be part of the results 2 days in the future, and those who be part of the results 3 days in the future? Then the results announced on different days would be completely independent of each other. I suspect that this is what he does.
I think you are right. But it is a fairly safe assumption that his voter demographics don't shift in a HUGE way from night to night, assuming he uses the same random dialing to generate info. Without a HUGE shift, it should be possible to make a reasonable guess at what his raw data shows. I can only base that on personal experience, and the fact that I am rarely surprised at the next days results based on my admitedly rough estimate of his data.