problem with x-axis labels after overlapping 2 box plots

I have the following code to compare two sets of box plots and there seems to be two misaligned x-axis labels after doing that! I have no idea what is causing this strange misalignment. Could the use of 'hold' be a problem in this case? Im using matlab 2014. Thanks.

x = rand(5,1);
y = rand(10,1);
z = rand(15,1);
Xpos = [2 5 12];

group = [repmat(Xpos(1), 5, 1); repmat(Xpos(2), 10, 1); repmat(Xpos(3), 15, 1)];
data=[x;y;z];

figure
boxplot(data, group,'positions', Xpos,'colors','k')

hold on

x =2+ rand(5,1);
y = 2+rand(10,1);
z = 2+rand(15,1);

group = [repmat(Xpos(1), 5, 1); repmat(Xpos(2), 10, 1); repmat(Xpos(3), 15, 1)];
data=[x;y;z];

boxplot(data, group,'positions', Xpos,'colors','r')

ylim([0 4])

[dpb]

Looks as expected on R2012b--no duplicate x-labels.

What's the 'renderer' setting for your case, just out of curiosity.

Is this reproducible on your system, I suppose? If so, I'd say it deserves a bug report to TMW at mathworks.com

[Chad Greene]

I'm also using 2012b and it looks fine for me too. I cycled through all the renderers and I can't repeat Dawn's duplicate labels.

[Star Strider]

I can’t reproduce it in R2014a. Looks fine — no duplicated or repeated x values.

To wit:

Although it doesn’t match Dawn’s plot. There may be something strange in Dawn’s workspace (workspace variables?) or in her code that we are not privy to.

The real Question is what Dawn did to get MATLAB mad at her?

[Dawn]

On first sight it may look fine but if I try to resize the plot by manually expanding it, the misalignment problem will show up. It happens all the time, even after clearing the workspace and pasting just the exact code that I posted above.

Would appreciate if the kind souls out there could let me know if to same thing happens to you after you resize your plots.

Answer 1

[Kelly Kearney]

The answer to the "why" is that boxplot, for reasons I've never been able to figure out, doesn't just change the tick positions and labels. Instead, it removes the xticks completely (or y-ticks, in the case of a horizontally-oriented boxplot) and replaces them with text annotation objects. And annotation objects are positioned relative to the figure, not the axis. Apparently slightly differently each time (possibly due to the fact that boxplot also, for reasons unknown, resizes axis, hence changing the figure-to-axis ratios between calls).

The easiest fix:

delete(findall(gcf, 'type', 'text'));
set(gca, 'xticklabelmode', 'auto', 'xtickmode', 'auto');

[Dawn]

It does the trick, thanks!

[dpb]

...boxplot, for reasons [unknown], doesn't just change the tick positions and labels. Instead, it removes the xticks ... and replaces them with text annotation objects

And how bizzaro!!! :( Seems worthy of bug report fer shure...at least an enhancement request to rid it of such foolishness. Who would ever think of doing such a thing????

[Kelly Kearney]

I think the original intent was to allow vertically-rotated tick labels (which is the default with the 'compact' option), but it's a pretty clunky implementation in my opinion.

I use boxplots quite a lot, and got really fed up with the heavy-handed axis changes, to the point that I wrote my own replacement:

boxplot2

Though my version handles multiple datasets a bit differently, and doesn't accept the Group input option.