Date:         Tue, 22 Oct 1996 15:46:42 -0400
Reply-To:     Ismail Parsa x6734 <sip@EPSILON.COM>
Sender:       "SAS(r) Discussion" <SAS-L@UGA.CC.UGA.EDU>
From:         Ismail Parsa x6734 <sip@EPSILON.COM>
Subject:      Re: new proc corr question
Comments: To: OLSON_K@A1.TCH.HARVARD.EDU

|>  From: "Karen L. Olson, Ph.D." <OLSON_K@A1.TCH.HARVARD.EDU>
|>  Subject:      new proc corr question
|>
|>  I now want to print the zillion correlations I am calculating.
|>  Thanks to Ian Whitlock, John Whittington, Jay Weedon, and others, the
|>  typing I had to do to run the analyses was greatly reduced.
|>
|>  On my final output, I want to print the correlation
|>  values for males, the significance of these values, correl for females,
|>  the significance of that, and the signif of the difference between
|>  males/females.
|>
|>  If I run proc corr by gender, I get my correl values and the exact
|>  p values.  However, when I save output to a file, there is no option to
|>  save the p values.  So I suppose I must calculate them.
|>
|>  Does anyone have a formula to calculate these exact p values?
 
Karen the following code will do what you are asking. Just
save your proc corr results into a data set named _corr_ and
let it work out the P values for you.
 
Regards.
 
*-----------------------------*
|        Ismail Parsa         |
|   Epsilon Data Management   |
|     50 Cambridge Street     |
|   Burlington MA 01803 USA   |
|                             |
| E-MAIL: sip@epsilon.com     |
| V-MAIL: (617) 273-0250*6734 |
|    FAX: (617) 272-8604      |
|                             |
|  The Usual Caveat Applies   |
*-----------------------------*
 
 
  data _cr_c (drop=_name_) ; set _corr_ ; if left(upcase(_TYPE_)) = "CORR" ; run ;
  data _cr_n (drop=_name_) ; set _corr_ ; if left(upcase(_TYPE_)) = "N"    ; run ;
 
  proc transpose data=_cr_c out=_cr_ct ( rename = ( col1 = _corr ) ) ; run ;
  proc transpose data=_cr_n out=_cr_nt ( rename = ( col1 = _n ) )    ; run ;
 
  data _pvals ; merge _cr_ct _cr_nt ; run ;
 
  data _pvals ( rename = ( _name_ = _VARS_ )
                keep   = depvar _name_ _corr _PvalC_ _pctmiss _N ) ;
   length depvar _name_ $8 ;
   set _pvals ;
 
   df       = _n - 2                        ;
   crnum    = (sqrt(_n - 2))*_corr          ;
   crden    = sqrt(1-(_corr*_corr))         ;
   crtrn    = crnum / crden                 ;
   _PvalC_  = (1-probt(abs(crtrn),df)) * 2  ;
   depvar   = upcase(left( "&depvar" ))     ;
 
   _pctmiss = 100 - ((_N / &__NOBS)*100)    ;
 
   format   _PvalC_ 7.5                     ;
 
  run ;