Hide non significant results

[trevismd]

I think this is the most frequently requested feature, already for statannot.

While the most "direct" way of conveying the number of tests performed is to plot the results for all of them, it would also be possible to explain all tests performed outside of the plotting area in order to have significant results stand out.

Those who seek to do that can do it rather easily already, with a few more steps. So the question is, should the package offer a simpler hide_non_signif_results option ?

Please react Up or Down if that matters at all to you :-)

[trevismd]

Related issues: #49, webermarcolivier/statannot#50

[sepro]

+1 for an option that would help creating a cleaner plot.

An alternative could be to specify if you want non significant results to 'show', 'mute' or 'hide'. Muting would be to show them with 50% opacity so they don't draw attention, but there still is a visual indication that test has been performed. Though that might open up an avenue for additional options how to style the muted comparisons ...

[annewiet]

Could someone please share sample code of the current workaround to hide non-significant results? Some of the comments to related issues are outdated I think.

[sepro]

Not a full example, but the code here is a good start. In the section Visualizing the results, if you filter non-significant results out of molten_df they won't show up in the plot.

[teanijarv]

+1 as well for this feature. Would be nice to just include an argument in the function to hide non-significant annotations or not.

If anyone has a current workaround this, please share the code. :)

[rkinet]

Yes +1 ! I'm also interested by this type of setting ! As @teanijarv said, the most user-friendly way should be including an argument ;)

[sepro]

I quickly coded up an example workaround how non-significant results could be removed for you and @teanijarv. You can find the example in this notebook at the very end.

Catch is that the significance testing needs to be done outside of statannotations which you probably need to do anyway when comparing multiple groups (see discussion about post-hoc tests).

[teanijarv]

Thanks @sepro . I already made my own workaround to my project, but good to have a well-explained notebook and solution for that for the next time. 👍

[sepro]

I'm messing with this in my own fork which allow you to do

...
annotator = Annotator(ax, pairs, data=df, x="group", y="value", order=order, show_non_significant=False)
...

and if the flag show_non_significant is set to False they will not be added to the plot anymore.

[sepro]

All automatic tests passed an it seems to work in the cases I tested!

PR #66 should add a flag to allow you to hide non-significant (defined as p >= 0.05) combinations.

[trevismd]

There seems to be quite the user demand for this feature, let's make that happen!

Thanks @sepro for giving the impulsion and contributing again!

[trevismd]

Now in the dev branch after work on #95.

[trevismd]

This is now included in v0.6!

Note that it is configured as a configuration parameter though (via .configure(hide_non_significant=True)).