Thursday, April 23, 2009

Tables versus graphs

Andrew Gelman argues that in scientific research "graphs are gimmicks, substituting fancy displays for careful analysis and rigorous reasoning". To him, the major problem is that graphs rarely show statistical significance, while a table "lays down your results, unadorned, for the readers--and, most importantly, scientific peers--to judge".

I am sympathetic to this view. After years in social networks, I have grown very tired of every researcher including a diagram of the network, which looks amazing but actually tells very little about the network.

But lately I have come to realize that its not enough to convey information, "unadorned, for the readers", because that is impossible. All science is interpretation, by the author first, then the reviewer, then the reader. A graphic can be an immediate, clear depiction of the data you need to convey.

Perhaps the greatest example of using a graph to depict data well is from Charles Joseph Minard of the losses suffered by Napoleon's army in the Russian campaign of 1812, which is hanging on the wall of my room:

This data could easily be put in a table, but it becomes so much more meaningful when displayed as a graphic.

I may be misunderstanding Gelman's point, but it seems that one shouldn't complain about graphs per se, but only the misuse of them. When done right, which is ultimately up to the reviewers and readers to decide, they can be far more informative than any table.


Chris Blattman said...

FYI, gelman was being sarcastic. He's a graph fanatic. Keep reading the piece.

Lee said...

Note the date on the post. It had me fooled too though.