What's fairly remarkable about this graph is that because the vertical dimension is so small, it looks rather as if Bush's approval has been fairly flat, if generally falling off a bit. One gets very little impression of the large swings in Bush's approval ratings prompted by events both unexpected (the attacks of 9/11) and all too expected (the invasion of Iraq), nor is there the strong sense that Bush's ratings fall off after each jump up, or that each jump up is smaller than the last, and that each decline is rather steep.

In other words, in extremis, people rally around their President, and then rather quickly are disappointed in his performance.

It's obviously not impossible to present the data in a way that makes this clear, because Professor Pollkatz does exactly that with a similar data set. (I've adjusted the horizontal scale to be approximately the same as the Roper graph):

In fact, the Roper graph itself could tell pretty much the same story, if they had chosen to use a more reasonable vertical scale -- here it is with the vertical dimension stretched 2.5 times, to be about the same as the Pollkatz graph:

It's still not as strong or clear a presentation of the data, but it's certainly less deceptive than the original.

I'm not accusing the Roper Center of deliberately skewing the graph to downplay Bush's falling approval ratings or minimizing the extent of his ups and downs (and downs, and downs), I think it's just sloppy presentation.

Addenda: I got curious about what the Pollkatz graph would look like if the vertical dimension of it was the same as the original Roper graph, so I squashed it, and got this:

Although even squashed like this the Pollkatz presentation is, in my opinion, superior, it's clear that it's the choice of vertical scale which is determinative, not the graphic design.