Remarks on Correlation Methods in Geophysics

Abstract

From consideration of several studies of the sampling properties of the product-moment correlation coefficient r it is concluded first that undue concern has been expressed for the problem of non-normality in correlation studies in geophysics, and second, that use of geo-physically adequate significance tests and confidence limits for r can almost always be achieved through use of the simple standard error of r. A growing tendency in climatology and hydrology to employ unnecessarily elaborate methods appears to stem from unrealistic emphasis which mathematical statisticians frequently place upon theoretical refinements, emphasis that loses sight of the limits of accuracy inherent in the very type of data usually subjected to statistical analysis.