Timeline for Explaining a boxplot and providing a reference in a technical paper
Current License: CC BY-SA 3.0
10 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Nov 13, 2013 at 19:55 | comment | added | Nick Cox | Absolutely no offence|offense inferred or taken; I am the one being candid. I have two copies of the original EDA book, but it appears to be out of print. Its many, many virtues don't include discussing why boxplots should be drawn one way rather than another, the kind of point that Tukey deliberately postponed for later accounts. Nor could Tukey in 1977 possibly cite later work, not that anyone is supposing otherwise. | |
| Nov 13, 2013 at 19:49 | comment | added | Andy W | @NickCox - I did not mean to offend by my suggestion and thank you for pointing to the reviews (I was unaware of them). Hadley's paper is clear in the abstract what it covers, and I would consider it larger in scope in terms of detailing extensions to bivariate displays. Your historical reviews are always excellent, and I didn't mean it as a substitute, only as a supplement. (IMO one would want to cite Tukey's EDA directly, but a copy of it is harder to get ahold of than either yours or Hadley's paper that are available online). | |
| Nov 13, 2013 at 16:50 | comment | added | Nick Cox | @Andy W The paper you cite appears to be dormant. I think it fair to mention that Hadley Wickham posted on GitHub not only the paper, but also two reviews from the American Statistician and two comments sent privately and independently to him by David Hoaglin and myself. See e.g. github.com/hadley/boxplots-paper/blob/master/reviews/… You can form your own judgements. | |
| Nov 13, 2013 at 16:09 | comment | added | whuber♦ | (+1) When explaining boxplots--which nowadays are almost always created by software--it is important to understand the software's conventions. Thus I would not recommend consulting Wikipedia, but instead look at your software manual. If that is not sufficient, some reverse-engineering (by constructing boxplots of tiny datasets where the computations are easily emulated by hand) usually reveals the details. | |
| Nov 13, 2013 at 13:25 | comment | added | Andy W | In addition to Nick's article I would suggest Hadley Wickham and Lisa Stryjewski's pre-print, 40 years of box-plots. | |
| Nov 13, 2013 at 9:14 | comment | added | Nick Cox | Explanation of boxplots is usually poor, a situation that perpetuates itself in several literatures. In particular, the Tukey convention that points are shown individually if and when more than 1.5 IQR from the nearer quartile is (a) not universal (b) not something that can be assumed obvious. stata-journal.com/sjpdf.html?articlenum=gr0039 is one reference discussing variations, which gives others. | |
| Nov 13, 2013 at 5:11 | comment | added | The Photon | Also there are different conventions for how to draw boxplots: What defines an outlier, how is the mean or standard deviation indicated (if they are) etc... Even in disciplines where boxplots are more common it would seem to be useful to explain your boxplots. | |
| Nov 13, 2013 at 3:42 | vote | accept | mrsteve | ||
| Nov 13, 2013 at 3:39 | history | edited | David Marx | CC BY-SA 3.0 |
added 21 characters in body
|
| Nov 13, 2013 at 3:29 | history | answered | David Marx | CC BY-SA 3.0 |