Date:         Thu, 28 Mar 2002 14:07:04 -0500
Reply-To:     Quentin McMullen <QuentinMcMullen@WESTAT.COM>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         Quentin McMullen <QuentinMcMullen@WESTAT.COM>
Subject:      Re: pairing variables for interaction variables
Content-Type: text/plain; charset="iso-8859-1"

Paula D wrote:
> I have 21 variables in a data set that I want to use to create new
> interaction variables. For example, to use var1 and war72 to create an
> interaction variable
>
> var1_war72=var1&war72;

Hi Paula,

Well, my SQL skills are pretty low, but I figure the only way to improve
them is to try a SQL solution every once in a while and then learn from
corrections/alternatives posted by others.  (e.g. it's no doubt crazy to run
two select statements on dictionary.vcolumns, just to grab the same thing
twice).

That said, the below seems to do what you want.  It creates all possible
2-way interaction terms from the variables in a dataset.  You'd need to add
to the where statement to exclude your outcome variable, etc.

And Peter Flom is of course correct, just because you *can* do this doesn't
mean you *should*.

*test data;

data a;
  array x{5};
run;

proc sql;
  select trim(name1)||'*'||name2 into :intlist separated by " "
  from
    (select name as name1
      from dictionary.columns
      where libname="WORK" and memname="A"
    )
    ,
    (select name as name2
      from dictionary.columns
      where libname="WORK" and memname="A"
    )
  ;
quit;

%put &intlist;

Kindly,
--Quentin