I’ve always subtracted 5 points from the Democratic advantage and usually come closer to the actual result than any of the pre-election polls. This guy, however, has closely examined a number of their structural flaws:
I have already written extensively about polling groups manipulation of demographic weights, so I will only summarize here that in addition to party affiliation, various polling groups this year have produced polls out of demographic balance with Census norms for urban/suburban/rural participation, minority race representation, age, employment status, and income range. It should not be difficult to imagine how these manipulations might invalidate the results published by the polling groups.
I have to admit, it always seems bizarre to me when a poll announces that in a recent polling of 850 Democrats and 600 Republicans, Obama has a 58-42 advantage. I mean, do they even have to ask the question to figure that one out? I understand the concept of estimating dynamic party identifications and electoral turnout, but it seems to me that given the weight of those estimates, the actual answers are irrelevant.
“In a poll of 140 black Democrats and one Republican who still supports Ron Paul, Obama is ahead 100-0. Looks like an epic Democratic landslide of unthinkable proportions, Katie!”
In summary, if McCain is within 5 points next month, he probably wins.