Comments on the interpretation of data from log normal distributions and definition of geometric standard deviation

Estimates of relative potency (commonly obtained as antilogs of log relative potencies) and the distribution of many pharmaceutical measurements are positively skewed and appear to follow a log normal distribution. The geometric mean arises naturally as a measure of location in these circumstances. There is not, however, a measure of dispersion which corresponds to the geometric mean and conflicting definitions have been used for the term geometric standard deviation. Examples of data for which the use of geometric means may be appropriate are given, together with comments on inferences which may be drawn regarding the population from which the data arise. It is suggested that the term geometric standard deviation leads to confusion and inappropriate inferences, and that it should therefore be avoided.