Is there a way to overlay score plot on loading plot in proc princomp?

[WarrenKuhfeld]

Princomp and prinqual give the same proportions every time I have ever tested them.

proc princcomp data=sashelp.heart;
   ods select eigenvalues;
   run;

ods graphics on;
ods html body='b.html';
proc prinqual data=sashelp.heart out=res scores mdpref;
   ods output mdprefplot=m;
      var ide(_numeric_);
   run;
ods html close;

proc contents varnum data=m; ods select position; run;

While I have not added it to the procedure at this juncture, I have grown fond of minimizing the data ink associated with the vectors.  See this blog.  It might not help much with hundreds of vectors.

https://blogs.sas.com/content/graphicallyspeaking/2017/06/22/advanced-ods-graphics-vector-plots-adju...

[variety]

Yeah, I got it why I am seeing two different results for princomp and prinqual.

For prinqual I am seeing this in log: The almost constant variable will be zeroed and excluded from analysis.

But for princomp, I don't see the above warning.

That might be it.

[WarrenKuhfeld]

https://blogs.sas.com/content/graphicallyspeaking/2017/11/02/advanced-ods-graphics-deeper-dive-docum...

Here is my blog that was inspired by this question.

[variety]

Hi @WarrenKuhfeld, Thanks so much for your reply and the blog post. For now, I am able to create the graph I am looking for using prinqual.

However, I still would like to learn how to overlay score plot on loading plot in princomp. As for my data prinqual is eliminating near zero variables and princomp is keeping the near zero variables, they are both showing two different results for my analysis. Also, for example: In PLS analysis, if I want to overlay, XSCORES on XLOADINGPLOT. Or XSCORES on XWEIGHTPLOT. It helps me with more tools to be able to modify my data. If there is any literature regarding this, please suggest me.

Thanks again for your help.

[WarrenKuhfeld · Accepted solution]

Since prinqual typically iterates, it tries to get rid of potentially problematic variables.  You can specify a really small singularity criterion as in this example.

data x;
   do i = 1 to 10;
      x1 = normal(7);
      x2 = normal(7);
      x3 = normal(7);
      x4 = normal(7);
      x5 = normal(7) * 1e-10;
      output;
      end;
   run;

proc prinqual data=x std;
   var ide(x:);
run;

proc prinqual data=x std singular=1e-50;
   var ide(x:);
run;

You can always output anything produced by any procedure and rearrange it in any way to make a graph.  Do an ODS OUTPUT on the data object that underlies prinqual's graph.  Then use princomp pieces to make something like that.  Again be aware that prinqual scales things in a nice way.  The documentation has details.