Fisher Z-Transformation

[Ano...@web.de]

I have to compare correlation coefficients, but first I must transfer
this coefficients in Fisher Z-Scores. I can't find such an option in
SPSS11. What should I do?

[Ano...@web.de]

[Marc Halbruegge]

> I have to compare correlation coefficients, but first I must transfer
> this coefficients in Fisher Z-Scores. I can't find such an option in
> SPSS11. What should I do?

Usually, I do this kind of stuff like this:
- generate the correlation matrix in SPSS
- copy the complete matrix to Excel
- use Excel-functions FISHER() and FISHERINV() to compute tests etc.

Greetings
Marc Halbruegge

[whea...@osu.edu]

I ran a Google search in this group on "fisher z-scores spss" and there
is a thread from 1997 that should answer your question too. Here's the
messge from Terry Taerum:

The following is code to test the difference in correlation
of marital with age for men versus women. It works since a correlation
is really a beta coefficient for z-scores and we are simply testing for
differences in slope. For your hypothesis you'll want to do a 1-tailed
test.

GET FILE='H:\Program Files\spss\GSS93 subset.sav'
/keep age marital sex .
compute nmissx=nmiss(age,marital,sex) .
select if nmissx eq 0 .
EXECUTE .
sort cases by sex .
split file by sex .
DESCRIPTIVES
VARIABLES=age marital /SAVE
/STATISTICS=MEAN STDDEV MIN MAX .
correlations vars=age marital .
split file off .
GLM
zage BY sex WITH zmarital
/METHOD = SSTYPE(3)
/INTERCEPT = INCLUDE
/CRITERIA = ALPHA(.05)
/print parameter
/DESIGN = zmarital sex sex*zmarital .

[Shu Fai CHEUNG]

In my honest opinion, the GLM approach should not work, although it may
lead to similar conclusion in some cases. The approach assumes that the
standard error in the GLM procedure is correct even for standardized
variables, when the standardization is based on sample standard
deviation. This should not be true. Otherwise, we don't need to have the
complicated formula for the standard error of a pearson correlation. We
can just standardize the variables and do a one-IV regression to get the
standard error. We don't even need the Fisher's z transformation, which
was derived from the complicated formula.

Just my two cents. If I am wrong, I would like to know.

By the way, I recall there is a MACRO for comparing correlation, which is
provided with SPSS and mentioned in the manual. Maybe located at the
program folder of SPSS and a subfolder. However, I don't have access to
SPSS at this moment. Is there anybody who can confirm or check?

--
Shu Fai Cheung
Faculty of Social Sciences and Humanities
University of Macau

"whea...@osu.edu" <whea...@osu.edu> wrote in
news:1143139777.8...@u72g2000cwu.googlegroups.com:

[jacc...@fiu.edu]

Check out the spss add-on at www.zumastat.com, It makes this easy.

[Shu Fai CHEUNG]

I found the entry for the MACRO for testing correlation difference.

It's SPSS Command Syntax Reference, Appendix D (Using the Macro
Facility), Example 2 (Testing Correlation Coefficients).

The exact page number may vary from version to version. However, I recall
seeing this entry in as early as SPSS 10.0

In some (or all) versions of SPSS, a soft copy of the syntax reference is
supplied with the program, and can be found in the program folder, or the
installation folder if you are using a network version.

Hope this helps.

"Ano...@web.de" <Ano...@web.de> wrote in news:1143071722.331348.220120
@u72g2000cwu.googlegroups.com:

> I have to compare correlation coefficients, but first I must transfer
> this coefficients in Fisher Z-Scores. I can't find such an option in
> SPSS11. What should I do?
>
>

--