Jaroslav Hlinka
2009-Jul-13 17:17 UTC
[R] testing equality of two dependent correlations + normality issue
Hi, I am turning to you with a (hopefully simple?) stats question. I would like to test equality of two correlation coefficients in a setting with three variables X,Y,Z, i.e. equality of r(X,Y) and r(Z,Y). I have found a formula to transform the "2 dependent correlations difference" to t-distribution with N-3 df: t = (rxy - rzy)* SQRT[{(n - 3)(1 + rxz)}/ {2(1 - rxy^2 - rxz^2 - rzy^2 + 2rxy*rxz*rzy)}] (Blalock, H., 1972. Social Statistics. NY: McGraw-Hill. Page 406-7). Am actually not sure whether this is exact or approximate (even given normality assumption, the Fisher's Z-transform which this is - I assume - based on, is approximate, right?). But to make it a bit more complicated, Shapiro-Wilks test of normality gives p=0.022 for variable X. Therefore assuming normality may not be safe (justifiable) at all? What do I do then? Do I report this test as "assymptotically valid", or do I run some other test? Any ideas? Many thanks in advance, Jaroslav -- View this message in context: http://www.nabble.com/testing-equality-of-two-dependent-correlations-%2B-normality-issue-tp24465768p24465768.html Sent from the R help mailing list archive at Nabble.com. [[alternative HTML version deleted]]