[VeritatisSplendor]

Mason-Dixon was the most accurate pollster in 2004, but SurveyUSA was 2nd most accurate:

[LS]

I'll take your word for it, but even then, you could be 92% accurate and be WRONG in all five or six truly competitive races.

For example, in 2002 I correctly called EVERY truly competitive senate race except for Thune; and in 2004 I correctly called every competitive senate race except for Colorado. So for Survey USA to have, say, McCain and Teddy K. winning, but miss, say, Thune's win or the GA race is kind of meaningless. I'd have to see how they were on races that were decided by less than 5% of the vote.