How to create florence nightingale graph: Coxcombs

[Arup Kumar Das]

Can some one suggest how to create florence nightingale graph: Coxcombs? I am attaching a picture of the graph for your reference.

[Nick Cox]

The only short answer I can give is "with difficulty". Stata's not really set up as a completely general graphics program. It could be done via lots and lots of calls to different twoway commands.

The rest of my answer is just a rant directed generally.

This graph is, in my view, one of the most over-praised graphs in the history of statistical graphics:

1. The circular format is presumably based on the idea that the variation is seasonal. Even if that were true, it is still hard work mentally comparing the same time of year for different years or different times of year for the same year. These problems are those which arise in interpreting pie charts and in comparing pie charts placed side by side.

2. The division into 12-month periods is driven just by the start of the data and is otherwise arbitrary.

3. In this case, seasonal variation is hardly dominant; the different numbers of casualties at different stages in the Crimean war is a major compounding factor, although more casualties died from disease than from wounds or injuries.

4. Interpreting amounts from areas of segments of annuli can hardly be said to easier or more effective than interpreting heights of curves.

5. A related problem is that if areas are used to represent amounts, people often still try to look at segments of radii; and vice versa. As areas scale as squares of lengths, the two are not equivalent.

6. Respect for this seems to be driven by respect (if not reverence) for the other achievements of the author, Florence Nightingale (often a bit exaggerated by the way, but that's another story) but that's no reason not to discuss it carefully and critically. (I don't know what she did to lose her capital letters, however.)

Incidentally, it's not even true that this was the first graph using a polar set-up, a miniature myth often repeated in the literature.

I discussed graphing this dataset at the 2013 London users' meeting. See e.g. http://www.stata.com/meeting/uk13/ab...s/uk13_cox.ppt but apologies if you don't have access to PowerPoint.

The most obvious time series graphs are not especially effective, as deaths from zymotic disease (roughly infections) swamp other causes for much or all of the time period. Conversely, logarithmic scale over-corrects. Two slightly non-standard graphs do a better job in my view, a "sparkline" representation in which each time series is shown within the same vertical space, and a square root scale. Both underline that the element of seasonality is secondary, as the winter of 1854-55 was not matched by that of 1855-56.

[Arup Kumar Das]

Thanks... this is useful...