= Comparing correlations in matrices from two subgroups  =

The following syntax produces individual group correlation matrices involving variables, x, saves the output to a SPSS data file and then compares two of the groups using Fisher's transformation (details and exmaples, for example in, Howell, (2002)).

In brief, we can transform a correlation, r, to ,r', where

$$\mbox{r' = 0.5} log_text{e} abs [ \frac{1+r}[1-r} } ]$$. 

Now, for a sample of size, N, the transformed correlation, r', follows a normal distribution with a mean of zero and a variance equal to 1/(N-3).

This means we can compare two correlations by comparing their transformed counterparts and so form a z score.

$$z = \frac{r'_text{1} - r'_text{2}}{\sqrt{1/(N_text{1}-3) + 1/(N_text{2}-3)}}$$

The macro below firstly uses the output management system (OMS) to obtain the correlations for two independent groups and then carries out the transformation and finally computes a z score which we can convert to p-values using other functions in SPSS. Any p-value less than 0.05 suggests there is evidence a pair of group correlations differ from one another.

{{{
* USING FISHER'S TRANSFORMATION FOR COMPARING TWO CORRELATIONS FROM TWO INDEPENDENT GROUPS

DEFINE !FISH ( GP !TOKENS(1)
             /  X !CMDEND).

SPLIT FILE BY !GP.

OMS 
/SELECT TABLES 
/IF SUBTYPES=['Correlations'] 
/DESTINATION FORMAT=SAV OUTFILE='c:\corrs.sav'
/COLUMNS SEQUENCE=[R3, R1]. 

CORRELATIONS
  /VARIABLES=!X
  /PRINT=TWOTAIL NOSIG
  /MISSING=PAIRWISE .

OMSEND.
!ENDDEFINE.

*
Specify group variable and variables in correlation matrix.

!FISH GP=SEX X=AGE EDUC PRESTG80.

* Edit the correlation data file and perform Fisher's test. 
* OMS creates names of form using the variable label    
* PearsonCorrelation_Var Label which are a bit cumbersome
* so you can rename them as below if you wish and remove 
* extraneous information.


GET FILE  ='C:\CORRS.SAV'.
SAVE OUTFILE='C:\CORRS2.SAV' 
 /DROP=Command_ to Var2 Sig.2tailed_Female Sig.2tailed_Male
 /RENAME=(PearsonCorrelation_Male PearsonCorrelation_Female N_Male N_Female = cm cf nm nf).
exe.
GET FILE='C:\CORRS2.SAV'.

SELECT IF (CF NE 1 AND CM NE 1).
EXE.

FORMAT CF (F5.2) CM (F5.2).

DEFINE !FISH2 ( GP1 !TOKENS(1)
             /  GP2 !TOKENS(1)
             /  NGP1 !TOKENS(1)
             /  NGP2 !TOKENS(1)).

COMPUTE Z= (0.5*LN(ABS((1+!GP1)/(1-!GP1))) - 0.5*LN(ABS((1+!GP2)/(1-!GP2))))  / SQRT( (1/(!NGP1-3)) + (1/(!NGP2-3)) ).
COMPUTE PZ= 2*CDFNORM(-ABS(Z)).
EXE.

SUMMARIZE
/TABLES=!GP1 !GP2 Z PZ
/FORMAT=LIST NOCASENUM NOTOTAL
/TITLE='CORRELATION COMPARISONS FOR MALES AND FEMALES'
/MISSING=VARIABLE
/CELLS=NONE.
!ENDDEFINE.

!FISH2 GP1=CM GP2=CF NGP1=NM NGP2=NF.
}}}

__Reference__

Howell DC (2002) Statistical methods for psychologists. Duxbury Press:Pacific Grove, CA