=========================================================================
Date:         Mon, 24 Jul 2006 15:03:11 -0500
Reply-To:     John Norton <jnorton@lumc.edu>
Sender:       "SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>
From:         John Norton <jnorton@lumc.edu>
Subject:      Re: Make bold or in color sign. correlations in output
Comments: To: jparry@spss.com
Content-Type: text/plain; charset=US-ASCII

Hi List:
 
I wanted to add a caveat (or a "gotcha") to James Parry's reference to the 'Correlations_Table_Correlations_Create' script.  This is quite a useful script, but when running correlations from command language and using the WITH statement, this script will still remove the diagonal and the upper half of the matrix, regardless of the WITH statement.
 
So, for example, if you execute correlations from the dialogs with variables q1 q2 and q3, you'll get a matrix with variables q1 q2 and q3 defining both the rows and the columns.  So, in the diagonal, you'll get perfect correlations - as variables tend to do with themselves - and in the upper right corner half of the table you'll get identical correlations as you would with the variable combinations in the lower left half of the matrix.  In this case, it makes sense that the diagonals of correlations of variables with themselves, and the redundant upper corner half of the matrix be removed.
 
But if you use command language with the WITH statement:
 
CORRELATIONS
  /VARIABLES=x1 x2 x3 WITH x4 x5 x6
  /PRINT=TWOTAIL NOSIG
  /MISSING=PAIRWISE .

..you'll get a matrix with variables x1 x2 and x3 defining the rows and variables x4 x5 and x6 defining the columns.  The script will still remove the diagonal and the upper right corner half of the matrix, regardless of the fact that in this case, there are no redundancies.
 
This is more of an annoyance and a speed bump than anything else (especially if/when attempting to demonstrate correlations in class, and after forgetting that you've already demonstrated this script and left it active).  Still, it's a "gotcha" to be aware of - it's always a good idea to monitor what you automate.  :-)
 
Best,
 
John Norton
Biostatistician
Oncology Institute
Loyola University Medical Center
 
(708) 327-3095
jnorton@lumc.edu
 
"Absence of evidence 
      is not evidence of absence"


>>> "Parry, James" <jparry@spss.com> 7/24/2006 2:15 PM >>>

Hi Martin (and list!),

There is a script that comes with SPSS called 'Correlations_Table_Correlations_Create' that 1) omits the top part of the diagonal of the correlation matrix and b) highlights in yellow any significant correlations (to .01, 2-tailed).

The way to activate the autoscript is to go to Edit...Options, choose the Scripts Tab and make sure to select the second script down 'Correlations_Table_Correlations_Create'. After selecting the script, make sure to press OK, and then run correlations.

Also, if you're interested in altering any of the parameters (say significance level to be highlighted, color of highlighting or font, etc.), or if you want to make the script work for Spearman's, here is a link to an earlier, detailed explanation by one of SPSS' Statisticians:

http://www.listserv.uga.edu/cgi-bin/wa?A2=ind9811&L=spssx-l&P=2898

If anyone happens to have a version of this script in another language (Python, VB, etc.) please feel free to submit it to: http://www.spss.com/devcentral/, our new meeting place for SPSS'ers interested in powerful statistical development.


Best,

James E. Parry