From FIVETHIRTYEIGHT.COM explaining this. Nate is a PhD pollster who does nothing but analyze polling data. He also doesn't "sell" his information to various clients like the for profit firms.
Oh yeah, and you don't want to get into a statistics swordfight with him either.
Matt Drudge is touting the results of a one-day sample in a Zogby poll, which apparently showed John McCain ahead by 1 point.
There are a couple of significant problems with this.
Firstly, there is a reason that pollsters include multiple days of interviewing in their tracking polls; a one-day sample is extremely volatile, and have very high margins for error.
Secondly, the Zogby polls have been particularly volatile, because he uses nonsensical party ID weightings, which mean that his weighting process involves making numbers doing naughty things that they usually don't like to do.
Thirdly, Zogby polls are generally a lagging rather than a leading indicator. This is because he splits his interviewing period over two days; most of the interviews that were conducted in this sample took place on Thursday night, with a few this afternoon. The reason this is significant is because lots of other pollsters were in the field on Thursday night, and most of them evidently showed good numbers for Obama, as he improved his standing in 6 of the 7 non-Zogby trackers.
Finally, there was no favorable news for McCain to drive these numbers. Polls don't move without a reason (or at least they don't move much).
FiveThirtyEight.com: Electoral Projections Done Right