[R-meta] Confidence Intervals of correlation coefficients

Guido Schwarzer @c @end|ng |rom |mb|@un|-|re|burg@de
Thu Jul 9 10:29:09 CEST 2020


The problem is that they use the correlation coefficient and its 
standard error directly which does not guarantee that the confidence 
limits fall into [-1, 1].

A better approach is to use Fisher's z transformation and to 
back-transform the results (with CI limits within the admissible range). 
This is the default in metacor() of R package meta. You can also use 
rma(..., measure = "ZCOR") in metafor, however, have to back-transform 
the results.

Best wishes, Guido

More information about the R-sig-meta-analysis mailing list