BUGFIX: Use correct terminology to avoid confusion

[hyiltiz]

trafficstars

Uses correct terminology to avoid confusion around whiskers vs. fences. Fences are defined as the bounds around the quartiles, whose length always equals k IQR (plotly adopts k = 1.5). Whiskers finds the data point closest to this bound instead. Plotly implements the latter, hence should use correct names.

https://stats.stackexchange.com/questions/149161/confused-by-location-of-fences-in-box-whisker-plots

[alexcjohnson]

Thanks @hyiltiz! Clearly we weren't aware of this convention when we created our implementation of the box plot, but it does look to me as though the distinction you're describing is used fairly consistently in the stats community. Unfortunately, this naming is part of the API, for example:

https://github.com/plotly/plotly.js/blob/f6c33bdb467458650d749fd7382d535c8f2aba62/src/traces/box/attributes.js#L113-L134

(oops part of the upperfence description still talks about the lower fence, cc @archmoj)

So it would be a breaking change to alter that now, and I don't think this is severe enough to warrant that big a step. I suggest we leave the API as is and address the confusion in the docs / descriptions.

The hover text labels we could certainly change, though I'll note that this is not so simple either since it's included in all the translations, so we would need to update those too, or add some custom code to fall back on the translation of upper fence if the dictionary hasn't been updated to include upper whisker.

[hyiltiz]

Thank you so much for the prompt feedback! I was confused about the hover text that I saw from ggplotly today, and it seemed the issue came from the js lib. I understand that introducing a breaking change to the API is undesirable, and would certainly be happy to accept that we change, as you suggested, all of the following:

• [ ] user facing UI elements such as hover text
• [ ] documentation for English and various other languages
• [ ] internal code that are not part of the external API (to avoid breaking changes)

As this warrants no urgent work, it would probably make sense to change all translations at once rather than going through the hassle of falling back to the incorrect terminology. This terminology can become especially confusing as the data becomes less Gaussian-like (e.g. a trimodal dataset, depending on the distance between the modes, drawing whiskers vs. fences would really make a large difference).

[archmoj]

(oops part of the upperfence description still talks about the lower fence, cc @archmoj)

Now fixed by #6834.