Barchart or Dotchart?

Which of the following two charts (both created with R) to you prefer? This dotchart:

Or this bar chart?

Andrew Gelman (who, incidentally, is speaking at the October NYC UseR meeting) prefers the dotchart prefers a line plot (update: see Gelman's comment, below), but personally I think the bar chart is more easily interpreted. What do you think? You can join the discussion at Decision Science News at the link below.

Update Sep 8: Prof. Gelman responds, with a detailed record leading to the correction above.

Decision Science News: Which chart is better?

Comments

[Barry]

I think there's something wrong with both. Dots only show position - people who read right-to-left might interpret the magnitude as being from the right.

The bars at least show magnitude irrespective of which way you normally read text, but since they don't start at 0 its difficult to interpret the relative magnitudes. At least with the dots you are forced to read the axis values.

[ugh, shouldnt have started this, gotta run]

[Andrew Gelman]

No! I prefer a line plot. See my comment on Dan's blog. I never said anything about a dot plot!

[David Smith]

Thanks for the clarification; I've updated the post above.

[GB]

It's not a fair comparison, since you've allowed one plot to use color and not the other.

Tell you what, send them in to a journal that prints only in black and white and compare them after that...

(I also object to barplots where the height of the plot doesn't represent a magnitude. If the bar is twice as long, it should represent twice as much of something.)

[Matt Shotwell]

The rationale (from the CJunk Charts post) for line plot versus bar plot or dot plot is good. But I was initially hung up on the connected dots, as this often implies interpolation.