One of the things I’ve noticed from working with scientists (of various background) is that they love tables full of means and standard deviations as a way of summarising the variability in some data or regression parameters. Andrew Gelman’s latest discussion of a paper makes the point that tables of numbers are awful and that a well made graphic does a good job of conveying the uncertainty. He refers in his comment to a paper he wrote, “Let’s Practice What We Preach: Turning Tables into Graphs” [1], which shows how graphs can be better at summarising variability, often in less space than a table. Another thing I really like about the paper is that it endorses the use of R/S/S+ for plotting and faults Excel for not offering enough control to the user (and it makes ugly graphs anyway).

I’m a big fan of using graphs because numbers don’t really mean that much to me, especially when dealing with things like splines and random walk models for non-linear function estimation. The UPTECH papers I’m writing on the fungal data and nanotracer measurements have a lot of graphs where previously there were tables or Excel plots which weren’t as easy to interpret. I’ve been spending quite a bit of time on them so that we can present to our readers, for example, just how different the means are in our hierarchical Bayesian model.

I think tables have a place and I use them in my own papers. I’m using a table in a spatial modelling paper to describe the prevailing winds and local geography at each of 13 measurement locations. There’s a map of the locations so that I don’t have to put things like “location” in the table. A list of features doesn’t translate as well to a plot as spatial locations do. I don’t think it’s appropriate to list row upon row of means, standard deviations, quantiles, etc. Long/wide tables of model fit criteria such as MSE, AIC, R^{2}, adjusted R^{2}, etc. are incredibly boring and do not scale well when you’re comparing more than, say, three models.

I think I might try to send this paper around my group as an attempt to convince them to abandon tables in favour of concise graphs. With the uptake of R among some of the more senior researchers/staff looking promising, I think it’s a message that might actually get some traction.

[1] Gelman, Andrew, Pasarica, C., and Dodhia, R. (2002). Let’s practice what we preach: turning tables into graphs. American Statistician 56, 121-130. [PDF]

### Like this:

Like Loading...

*Related*

Rohan J“One picture is worth ten thousand words” – Fredrick R. Barnard.

So, that translates into “One graph is worth five hundred tables”.

Sam CliffordPost authorI’ve always heard it as “a picture is worth a thousand words”. Maybe words in physics are less valuable than in statistics. :P

Rohan JThat’s what I thought too, until I checked my Concise Oxford Dictionary of Quotations … :-)