We are preparing a manuscript and the editor asked us to convert a figure with a boxplot into a table "because of the more exact data content". While I think boxplots are pretty decent in revealing something about the data, I was wondering what you guys think about this? Do you (often) opt for tables over boxplots for presenting your data?
$\begingroup$
$\endgroup$
asked Dec 4, 2010 at 9:05
Add a comment
|
2 Answers
$\begingroup$
$\endgroup$
I tend to think that boxplots will convey more effective information if there are numerous empirical distributions that you want to summarize into a single figure. If you only have two or three groups, editors may ask you to provide numerical summaries instead, either because it is more suitable for the journal policy, or because readers won't gain much insight into the data from a figure. If you provide the three quartiles, range, and optionally the mean $\pm$ SD, then an advertised reader should have a clear idea of the shape of the distribution (symmetry, presence of outlying values, etc.).
I would suggest two critical reviews by Andrew Gelman (the first goes the other way around, but still it provides insightful ideas):
- Gelman, A, Pasarica, C, and Dodhia, R. Let's practice what we preach. The American Statistician (2002) 56(2): 121-130.
- Gelman, A. Why Tables are Really Much Better than Graphs. (also discussed on his blog)
$\begingroup$
$\endgroup$
It depends on your objective, if it is a quick visualization I will stick with boxplot but if it is a more detailed examination I would stay with the data
answered Dec 4, 2010 at 9:22