topic Lognormal scale boxplot in Graphics Programming

Lognormal scale boxplot

PGStats — Wed, 28 Aug 2013 16:20:30 GMT

Hi all,

This SGPLOT-HBOX graph is almost what I want

It shows the distribution of a concentration on a log10 scale. Means and far outliers are also determined on a log scale. What I would like is the X axis to show original units instead of log units. I need a solution that will accommodate BY processing with varying ranges on the X axis.

Thanks.

Re: Lognormal scale boxplot

DanH_sas — Wed, 28 Aug 2013 16:29:30 GMT

Can you post the SGPLOT code you're using? What logstyle are you using?

Re: Lognormal scale boxplot

PGStats — Wed, 28 Aug 2013 17:08:04 GMT

Here is the code

proc sgplot data=metDet;

by element;

hbox logvaleur / category=grpOrganisme datalabel=site labelfar;

*xaxis type=log logbase=10 label="Teneur (mg/kg)";

xaxis label="Log(Teneur (mg/kg))" integer;

yaxis display=(nolabel);

run;

If I use the commented out XAXIS and valeur instead of logvaleur, the mean and outliers (and whiskers, I think) will be determined on a linear scale, which is not what I want.

Message was edited by: PG Added the change of variable required to go to a linear scale.

Re: Lognormal scale boxplot

DanH_sas — Wed, 28 Aug 2013 20:08:05 GMT

Here is a trick that might work for you:

1. Create a user-defined format that maps the log tick values back to the original values. Create a mapping that will encompass the total range of data across the BY-groups. The INTEGER option that you're using should make that mapping easier to create.

2. Use the TICKVALUEFORMAT option to assign the format to the X-axis.

Hope that helps!

Dan

Re: Lognormal scale boxplot

PGStats — Wed, 28 Aug 2013 21:19:49 GMT

But that's cheating! OK, I'll try it . Thanks Dan. - PG

Re: Lognormal scale boxplot

PGStats — Thu, 29 Aug 2013 02:58:22 GMT

It worked nicely in most cases. It wouldn't look good for cases where the X range is too short.

Thanks again!