How to read a poll

I suppose it doesn't really matter, since public opinion is what it is, and polls are just a clumsy way to measure it, but it is absurd that this headline appeared in so many papers today: Newsday.com - President's Job Approval Rating Rebounds, followed by this lede:

President George W. Bush's job approval rating, which had slumped in several recent polls, has bounced back to 56 percent in a CNN-USA Today-Gallup poll released yesterday.

Here's political analyst Charlie Cook's take, from the very best poll, also today:

Since Labor Day, the decline looks more like a real drop than a
settling, and the triple issues of the economy, the deficit and Iraq
have become conjoined. An Ipsos/Cook Political Report poll taken last
Tuesday through Thursday of 787 registered voters showed Bush's approval
rating had dropped four points from three weeks ago, from 55 percent to
51 percent. That was six points lower than the 57 percent approval
rating seen in both the months of July and August.

It's this simple: The CNN/Gallup approval numbers have always been higher than other polls, by five or six points. I'll leave it to Ruy Teixera to explain exactly why, although I'll note that the reason is usually not bias, but a simple methodological difference. (This beautiful chart at Pollkatz shows the various polls -- the pink diamonds near the top each week are the CNN-Gallup numbers.) The 50% approval a week or so ago was an anomaly in the Gallup poll, and everyone knew it. Having Bush back at 56% approval is just a return to the normal trend for this Bush-tilting poll, and it's still lower than his approval in that particular poll at any time between Sept 11, 2001 and August 2003.