[Nuts and bolts of percentiles]

Decisions regarding diagnosis and treatment of medical conditions are often based on reference values, and these are frequently based on percentiles. We have studied the impact of statistical methods and software on estimates of percentiles.

Reference values for semen parameters have been based on measured values in fertile men. We have reanalysed our previously published data with different statistical methods for estimation of percentiles. These methods are available in the programme R through the quantile function and include those used in standard programmes such as Minitab, SAS and SPSS.

Analyses of semen parameters by different non-parametric methods have revealed considerable differences in 5 and 10 percentiles for semen concentration. The differences were substantially smaller for semen motility. Our results imply that reference values should be accompanied by a detailed explanation of the statistical methods used for the estimation of percentiles. In a clinical setting, the choice of method may determine if a test result close to the reference values is considered pathological or not. It is therefore vital to realise that different methods may produce different results. Moreover, it may be difficult to compare reference values reported by different laboratories if the statistical methods differ.