Well, of course a call-in talk show represents "a sample." It's just not a statistically significant sample and, therefore, worthless as a basis of drawing correct conclusions.

The nature of a talk show inherently has a biased database and is in no way a representative sample. For instance, what percentage of Rush Limbaugh's listeners would characterize themselves as "liberals"?

What percentage of Air America Radio listeners would characterize themselves as "conservatives?"

Professional pollsters, those who make a living on trying to ferret out ACCURATE results, know that there is actually a SCIENCE to these sort of things - and call-in talk shows are certainly not scientific.