DATA Step, Macro, Functions and more

How to create a decompostion tree?

Reply
Occasional Contributor t30
Occasional Contributor
Posts: 9

How to create a decompostion tree?

www.bportugal.pt/PAS/en/help/image/decompTree.gif

 

Example is as above.

 

Often times proc freq with multiple variables become illisible.

 

Is there any way to recreate something as above?Capture.PNG

PROC Star
Posts: 1,561

Re: How to create a decompostion tree?

This would be a nice VA object.

I don't think there is a built-in way to do this with SAS at the monent though.

Super User
Posts: 10,500

Re: How to create a decompostion tree?

The graphic part I'm not going to guess at.

You could create a multilable format and use Proc tabulate to do the tables.

 

Below are a couple of examples with different orders of multilable formats for the same sort of data and options in Proc Tabulate to get the groups displayed. Title statements show the options used. Note that the Mulilable formats only work in a few procedures: Procs Means, Summary and Tabulate.

 

/* To demonstrate that the order of definition affects appearance in
   multilabel formats. Also appearance options to show the spaces to
   get the indent as desired and fix column widths.
   And investigate whether class level format based style overrides work
with MLF in proc tabulate.  Result: NO.
*/
proc format library=work;
value accidentl (multilabel notsorted)
1-5 = 'Accidents'
1-3 = ' Transport accidents'
1 = '   Motor vehicle accidents'
2 = '   Water, air, and space'
3 = '   Other land transport accidents'
4-5 = ' Nontransport accidents'
5 = '   Fishing'
;
value accidentr (multilabel notsorted)
1-5 = 'Accidents'
1-3 = ' Transport accidents'
4-5 = ' Nontransport accidents'
1 = '   Motor vehicle accidents'
2 = '   Water, air, and space'
3 = '   Other land transport accidents'
5 = '   Fishing'
;
value accidents (multilabel notsorted)
1 = '   Motor vehicle accidents'
2 = '   Water, air, and space'
3 = '   Other land transport accidents'
5 = '   Fishing'
1-5 = 'Accidents'
1-3 = ' Transport accidents'
4-5 = ' Nontransport accidents'
;
value mf
1 = "Male"
2 = "Female"
;
value mycolor
1='white'
2='red'
3='green'
4='blue'
5='orange'
6='purple'
;
value accsimple
1 = 'Motor vehicle accidents'
2 = 'Water, air, and space'
3 = 'Other land transport accidents'
4 = 'Nontransport accidents'
5 = 'Fishing'
;
value MyColorl (multilabel notsorted)
1-5 = 'white'
1-3 = 'red'
1 = 'blue'
2 = 'orange'
3 = 'pink'
4-5 = 'purple'
5 = 'black'
;
run;

/* populate a dataset to display */
/* This specifically does NOT generate any data for FISHING above*/
/* to display the behavior of the options below in those cases. */
data junk; 
do i=1 to 50;
type = round(4*ranuni(1234)+.5);
sex = round(2*ranuni(3455)+.5);
output;
end; 
run;


/* Notice that before we get here the data is NOT sorted */
/* in any manner!!!! */
title 'Option Order=Data Preloadfmt Printmiss Misstext=0 format accidentl';
proc tabulate data=junk order=data ;
   class type / mlf PRELOADFMT;
   /*ASIS preserves the leading spaces in the format, Cellwidth here is optional
     for this demo. These will ONLY affect ODS output such as HTML, RTF or PDF
     not the OUTPUT window
  */
   classlev type/ style=[asis=on cellwidth=1.5in];
   class sex;
   /* the formatted values of SEX normally take up different lengths, this fixes
      the column widths to be the same for both headers. May cause interesting
      appearances with large numbers of displayed digits if requested in formats*/
   classlev sex/ style=[cellwidth=.5in];
   /* to get the ALL to have the same width as SEX need to specify this way*/
   table type=' ', all='Total'*{style=[cellwidth=.5in]}*n=' '*f=comma9. sex=' '*n=' '*f=comma8.
   / printmiss misstext='0';
   format type accidentl. sex mf.;
run;title;

title 'Option simple format and classlev background color';
proc tabulate data=junk order=data ;
   class type / ;
   /*ASIS preserves the leading spaces in the format, Cellwidth here is optional
     for this demo. These will ONLY affect ODS output such as HTML, RTF or PDF
     not the OUTPUT window
  */
   classlev type/ style=[asis=on cellwidth=1.5in background=mycolor.];
   class sex;
   /* the formatted values of SEX normally take up different lengths, this fixes
      the column widths to be the same for both headers. May cause interesting
      appearances with large numbers of displayed digits if requested in formats*/
   classlev sex/ style=[cellwidth=.5in];
   /* to get the ALL to have the same width as SEX need to specify this way*/
   table type=' ', all='Total'*{style=[cellwidth=.5in]}*n=' '*f=comma9. sex=' '*n=' '*f=comma8.
   / row=float misstext='0';
   format type accsimple. sex mf.;
run;title;

title 'Option Order=Data Preloadfmt Printmiss Misstext=0 format accidentl classlev background';
proc tabulate data=junk order=data;
   class type / mlf PRELOADFMT;
   classlev type/ style=[asis=on cellwidth=1.5in background=mycolor.];
   class sex;
   classlev sex/ style=[cellwidth=.5in];
   table type=' ', all='Total'*{style=[cellwidth=.5in]}*n=' '*f=comma9. sex=' '*n=' '*f=comma8.
   / printmiss misstext='0';
   format type accidentl. sex mf.;
run;title;
title 'Option Order=Data Preloadfmt Printmiss Misstext=0 format accidents';
proc tabulate data=junk order=data;
   class type / mlf PRELOADFMT;
   classlev type/ style=[asis=on cellwidth=1.5in];
   class sex;
   classlev sex/ style=[cellwidth=.5in];
   table type=' ', all='Total'*{style=[cellwidth=.5in]}*n=' '*f=comma9. sex=' '*n=' '*f=comma8.
   / printmiss misstext='0';
   format type accidents. sex mf.;
run;title;
 

SAS Super FREQ
Posts: 3,478

Re: How to create a decompostion tree?

Are you asking about how to summarize the data to obtains the counts/percentages? Or are you asking about how to visualize a two-way crosstabulation? PROC FREQ can do both. I recommend the mosaic plot for many two-way tables, although since 94% of these data are bikes, it might be hard to label the other cells in the mosaic plot. PROC FREQ can create other graphs, such as a stacked bar chart.or a modified stacked bar chart that shows relative percentages.

Occasional Contributor t30
Occasional Contributor
Posts: 9

Re: How to create a decompostion tree?

Hi Rick_SAS, I regularly have a 4 level of trees, with each level having 2 or more items.

I think mosaic plot works great for 2 levels, not so when we have more.

PROC FREQ already gives the counts/percentages, which is great. But if you have var1*var2*var3, the PROC FREQ table can be a bit too lengthy to read from.
Ask a Question
Discussion stats
  • 4 replies
  • 91 views
  • 0 likes
  • 4 in conversation