Methodological issues in the analysis of human sperm concentration data

Abstract

We examined two methodological issues in the analysis of sperm concentration data using a large database of sperm concentrations in healthy men that were collected at the University of Washington. We showed that the raw data were skewed and that log transformation should be used to assure that the data meet the assumptions underlying most statistical estimation and testing procedures. We also addressed the issue of the great variability in sperm concentrations within a single individual and the necessity and utility of multiple sampling to reduce variance. We conclude that log-transformed data should be used for statistical analysis of sperm concentration and recommend that such analyses be based on the geometric mean of several samples from each subject to reduce variability, increase accuracy of estimation, and improve statistical power. This is particularly important when the objective is to detect small but important differences or subtle effects.