Cleveland vs Huff

My post from two days ago reminds me of something.

In How to Lie with Statistics, Darrell Huff tells us that an honest graph will include zeros on its axes. He gives the following example.

The graph on the left makes “an increase of under four per cent look like more than 400”, while the graph on the right is “honest”.

William Cleveland responds to this example in The Elements of Graphing Data. He points out that adding the zero has destroyed the resolution of the graph. It’s silly to insist on including zero; we should expect the viewer to notice the tick labels and have a modicum of quantitative sense.

Seriously, if a viewer is overimpressed by the graph on the left, then the same viewer will probably be underimpressed by the graph on the right.

So, I agree with Cleveland.  But I would also say (and this is where the previous post comes in), if the viewer ought to be able to gauge per cent change, then show per cent change:

(By the way, I think the title “How to Lie with Statistics” wasn’t a good choice.  That is, unless Huff wanted people to think that’s what statistics are for, because I’m sure it’s had that effect.  Overall, though, it’s a good book, and has at least one good message: Don’t be content with less than a full explanation.)