Hi Holger, at least to my knowledge, this is not implemented in SPSS. The (german) textbook by Bortz (1989, 3rd ed., p. 264) states, that the difference between two Fisher-Z-transformed correlations is normally distributed, and thus the Ho r1 - r2 = 0 can be tested using the following tests statistic:
z = (Z1-Z2)/sigma(Z1-Z2) where sigma(Z1-Z2) = sqrt(1/(n1- 3) + 1/(n2 - 3)). (unfortunately, Bortz gives no reference). However, I implemented this is in a SPSS macro (cor_dif2.sps), which may be donwloaded from the spss-macro-section of my hp: http://www.allg-psych.uni-koeln.de/naumann/makros.htm (unfortunately, there is no readme-file yet, i hope i'll find the time to write one soon ...) Hope this helps, Johannes Bortz, J. (1989). Statistik f�r Sozialwissenschaftler. Berlin: Springer. (3rd ed.) [Statistics for the social sciences] -- Johannes Naumann (Dipl.-Psych.) Universit�t zu K�ln, Psychologisches Institut Herbert-Lewin-Str. 2, D-50931 Koeln, Germany Tel. ++49(0)221/470-3848 / Fax -5002 http://www.allg-psych.uni-koeln.de/naumann/index.htm . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
