How to put a boxplot on a scatterplot/histogram?

I'm trying to put a boxplot over the data points in this scatterplot but can't find anything. This is the code to setup the table for the scatterhistogram

tbl = table(swine, cc, depth);
s = scatterhistogram(tbl, 'swine', 'cc', 'GroupVariable', 'depth');

The scatterplot is attached. Many thanks

[dpb]

So, what did you try? Won't simply

hold on
boxplot(...)

where the ellipses are whatever variable is that want the boxplot for do it?

[Adam Danz]

Actually it's not that easy with scatterhistograms. They are their own class and do not provide access to the axes at all.

[dpb]

Ah, yes...another case of TMW getting far too clever for their own good. :(

[Adam Danz]

Yeah, I'm not sure I understand the reasoning behind the family of high-level, standalone visualizations such as scatterhistogram, that mirror other visualization functions but without the same level of customizability.

To make things even more confusing, there are inconsistencies in the param names between the two functions (ie, nbins vs NumBins, which appear to do the same thing).

TMW have made some improvements to the standalone functions, though. I submitted a feature request earlier this year because scatterhistogram wouldn't allow us to add anything to the axes such as yline(),text(), or even a marker to indicate mean/median values using plot(). boxchart() is another example of a standalone version of boxplot() but I recall being able to make these modifications in boxchart. It looks like theses capabilities have been added to scatterhistogram in the r2020b release.

I'm guessing that their motovation is to streamline the use of these functions without the need to manipulate handle properties but if you compare the inputs between the standalone versions and the customizable versions, they aren't that much different. The standalone versons also give TMW greater flexibility in updating the functions without worrying about backward compatibility associated with external property manipulation. Except for the most basic use cases, I bet a lot of newer users eventually migrate their visualization from the standalone versions to the customizable versions as in this thread.

[dpb]

Yeah, seems like they waste a LOT of time on such trivia and neglect far more significant improvements. Meanwhile the code bloat just continues to mutiply.

It certainly is not clear who's driving the ship and where they think they're going.

Answer 1

[Adam Danz]

scatterhistogram() is its own class and does not provide access to its axes. Instead, use the older scatterhist(). It's output is a 1x3 vector of axis handles where the first handle is the axis to the scatter plot.

h = scatterhist(___);
hold(h(1), 'on')
boxplot(___)

However, the marginal histograms are normalized and the boxplots will likely be at a much larger scale than the histograms. To see the axis scale, turn on the axes using,

axis(h(3), 'on')

I suggest putting the boxplots elsewhere, perhaps in their own axes or within the scatter axes.

[Saar Peles]

Is it possible to also just put points onto a boxplot? Maybe that would be easier

[Saar Peles]

I tried something similar but the boxplot wouldn't stay. I asked a previous question about it here: https://www.mathworks.com/matlabcentral/answers/616608-scatterhist-function-not-allowing-for-boxplot-on-the-y-axis-to-stay?s_tid=srchtitle

[Adam Danz]

"Is it possible to also just put points onto a boxplot?"

Yes. See this demo:

https://www.mathworks.com/matlabcentral/answers/459779-display-data-points-in-boxplots#answer_373270