Charles Franklin at Pollster.com responds today to reader claims that Rasmussen skews poll averages of President Obama's approval rating by turning in numbers that are consistently 2-3 points below the average.
It's true--Rasmussen's daily polls are low (see the graph below), but Franklin points out that Gallup's daily polls are also consistently high, and the two balance each other. Pollster.com's overall average, meanwhile, is remarkably close to the average of non-daily polls, published by a heterogenous group of 23 different firms.
So its just a matter, as Franklin says, of picking one's poison. More broadly, it raises an epistemological question about the nature of polls and skewed data: how does one know that polls like Rasmussen's are low, or that Gallup's are high? One would have to know Obama's true approval rating, in a rationalistic sense that doesn't mesh with the empirical process of poll-taking.
In other words, you can exclude a poll that's lower than the average because you think it's low, but that's exactly what you're doing--selectively altering the data and driving up the trend.