Description
The standard formula we use for the length of box plot whiskers is 1.5 * IQR (interquartile range, ie the third quartile minus the first quartile). This is based on comparing your distribution to a normal distribution, where any data beyond those whiskers is considered an outlier. However, if you're displaying your data on a log scale (and sometimes even if you're not) a log-normal distribution is likely more appropriate, in which case the whisker locations should be calculated based on the IQR in log units. This brings both fences up, in particular preventing the lower fence from ever going negative ie an infinitely long whisker on a log scale.
I'd propose a new attribute for box traces, something like distribution: 'normal' | 'log-normal' | 'auto'
where 'auto'
uses 'log-normal'
when displayed on a log axis, otherwise 'normal'
. I'll be happy to submit a PR to add this functionality.