Basic:fisher ' s transform

Source: Http://bbs.chinahrd.net/thread-709742-1-1.html,Kenneth's answer.

z = 0.5 * ln [(1+r)/(1-R)]
"C0 Z2 N j/p; L3 t0 O
The transformed distribution is generally normal. and its standard deviation is not affected by the size of the overall correlation coefficient, but becomes a simple function of the sample number 1/sqrt (N-3), N is the size of the sample.

Without conversion, the sampling distribution of correlation coefficients is biased, not normal, and the larger the correlation coefficient, the larger the standard error of sampling distribution. This is a fact that is very bad at verifying Ho: the overall correlation coefficient = 0 o'clock. So, to turn the correlation into Fisher Z. Then it's much simpler to verify that the overall correlation is 0. (J ' N% K3 b7 a% u '? 8 m
But if you want to draw a confidence interval (such as an estimate of a standard deviation), calculate the Fisher Z +-SD and turn it into a simple correlation coefficient r.

Basic:fisher ' s transform