Is there a way around to make a box plot more nicely visuable when having extreme values that compresses the boxes and whiskers?

[Sigve Aarflot]

Dear users.

I am trying to present return on assets and loan losses of banks in a box plot over a 10 year period. The problem is however that due to extreme values on some observations in my data, the actual whiskers and boxes are compressed such that it seems they lie on a flat line over the sample period. This box plot is not presentable in my opinion. I recon that changing the scale of the y-axis is limited to the most extreme values and that it does not allow removing the most extreme outliers by scaling down the y-axis. Is there an alternative solution for my problem, or a way around to fix it to make the box plot more nicely visuable ? Thanks in advance !

The attached file shows the box plot of return and loanloss over the 10 year period.

[Jorrit Gosens]

Consider the nooutsides option

[Nick Cox]

If boxes seem too small, you might consider using a transformed scale.

An important nuance is that identification of whiskers doesn't commute with nonlinear transformations. For that reason and others, forms of box plots based directly on percentiles (quantiles) have an edge over those codified in -graph box- or -graph hbox-.

[Sigve Aarflot]

Jorrit, appreciate the quick feedback!

Nooutsides certainly did the trick, so thanks. However, shouldn't the box plot represented in a descriptive statistic section present all observations one uses in the analysis, or is it sufficient to present the observations between the interval of 25th and 75th percentile and put a comment under?

[Nick Cox]

I'd turn #4 around. If showing the middle half of the data, only, satisfies research aims and reviewers, so be it.

[Sigve Aarflot]

Nick, thanks for the feedback!

I will consider what is most appropriate.

[Nick Cox]

Earlier I was in a meeting and couldn't see your graph or post attachments properly.

Reminder: you are asked not to show .gph attachments. This is explained in the FAQ Advice everyone is asked to read before posting.

Here it is more visibly for many.

For variables like this which can be positive, zero or negative, possible transformations are

cube root

neglog := sign() * log(1 + abs())

asinh