topic Re: Boxplots With Varying Endpoints, Boundaries, Fences, or Percentiles in Graphics Programming

Boxplots With Varying Endpoints, Boundaries, Fences, or Percentiles

Zachary — Thu, 05 Feb 2015 14:23:35 GMT

I am about to start studying the code for building boxplots with SAS. I noticed there are many ways or procedures to build them. Which one is best for the following?

I wish to do a box plot for 10 groups of data, side-by-side. I thijnk that will be easy enough. I am not really interested in the fanciness of the graphics, but I am sure that will come in time.

I also wish to vary the end points for my whiskers or outliers. For example, I believe the standard box plot pushes the outer fences to 1.5 X the IQR. I would like the option of seeing the data at 2 X IQR range or 3 X IQR as well. Or, maybe perhaps specify the percentiles.

Thank you very much in advance.

Re: Boxplots With Varying Endpoints, Boundaries, Fences, or Percentiles

Reeza — Thu, 05 Feb 2015 14:36:26 GMT

If you want to customize it use GTL with WHISKERPERCENTILE which is an option in SAS 9.4 at least.

I'm not sure about 2XIQR or 3XIQR. Would also have to be careful that wasn't over the max/min of the data.

Re: Boxplots With Varying Endpoints, Boundaries, Fences, or Percentiles

Jay54 — Thu, 05 Feb 2015 14:55:49 GMT

WHISKERPCT is also supported by SGPLOT with SAS 9.4M2.

With earlier releases, we have used HighLow plot do do the same:

http://blogs.sas.com/content/graphicallyspeaking/2013/03/24/custom-box-plots/

Re: Boxplots With Varying Endpoints, Boundaries, Fences, or Percentiles

Zachary — Thu, 05 Feb 2015 15:41:50 GMT

That is great information from both of you guys! Thank you so much.

Now the problem I am dealing with is my scatterplots are skewed in a major way. Wihin each of the 10 categories the majority of the points are zero - no longer thinking boxplots would be terribly conducive to understanding such a skewed distribution.

Any other suggestions to graphically see what is going on in a snapshot? If I were to just send out my percentiles an example for one of my categories could be:

P1 0

P5 0

P10 0

P25 0

P50 0

P75 6

P90 22

P95 47

P99 128

Can anyone think of a way to adjust such scatterplots in these situations? Perhaps put a condition in that the value needs to be > 0 - but then I am steering away from the original question.

Thank you.

Re: Boxplots With Varying Endpoints, Boundaries, Fences, or Percentiles

Jay54 — Fri, 06 Feb 2015 17:08:40 GMT

With SAS 9.4M2, you can overlay scatter plot on a BoxPlot. Would that help?

Here, for many observations (chosen randomly), I made mpg_city=missing, and mpg=0 for the same.

Then I plotted a box plot of mpg_city, overlaid with a scatter plot of mpg with JITTER.

Does this help?

data cars;

set sashelp.cars;

if ranuni(2) < 0.7 then do;

mpg=0;

mpg_city=.;

end;

run;

proc sgplot data=cars noautolegend;

vbox mpg_city / category=type;

scatter x=type y=mpg / jitter;

run;

Re: Boxplots With Varying Endpoints, Boundaries, Fences, or Percentiles

Zachary — Fri, 06 Feb 2015 21:25:11 GMT

Thank you Sanjay. My chart is similar to yours - but my boxes do not show up at all given the quality of the data I am working with.