Interesting. In an earlier<$Ad$> post
I said that I didn't put that much stock in polls by SurveyUSA because they don't use people to conduct their polls, but an automated response system.
A lot of people in the public opinion and poll business share my skepticism. But an academic public opinion researcher, Joel David Bloom, did a study
of their results and found them about as reliable as the rest. Here's the key graf ...
Looking at SurveyUSA, they too stuck with Carnahan in Missouri, and had both Carolinas much tighter than they actually were. But they were right in Colorado, and had the margins closer elsewhere. Interestingly, they were not in as many close states as Zogby, so the fact that 85% of their polls had the correct winner should not be taken too far. Unlike Zogby, or nonpartisan pollsters in general, SurveyUSA did show a net pro-Democratic bias of around 2 points, but by every other measure they performed as well as or better than other nonpartisan firms. Thus, as much as academic survey researchers may have wished to see SurveyUSA under-perform the field, they clearly did not, and may have actually done better than average.
I remain skeptical. But those are interesting numbers.