Rasmussen polls registered voters (how he confirms this is anyone’s guess - the honor system?) and has been pretty close on most races, much more than the LameStream Media polls, which ask any illegal or felon what they think.
Well, sure, he has to find "registered" voters first, to separate them from "all adults." But then he filters for likely voters. And his final results are always reported in terms of likely voters, whereas almost all other surveys use registered voters right up until a few weeks before the election -- at which point many of them switch from registered to likely.
Now the methods that Ras uses to filter out the "registered" and "likely" categories must be proprietary. I'm sure he doesn't want the competition to know how he achieves his results. But achieve them he does, and he clearly has the best long-run record for accuracy.
The above having been said, some member of the regular anti-Ras crowd (actually, just a handful) here on FR will probably pop up now and tell us how the current poll is just another instance of Rasmussen's trying to dupe us. Ho hum. What else is new?