Editorial:‘Robust’ statistical methods

Abstract

Investigations on the validity of the statistical analyses presented in biomedical journals, even those that are well reputed, have found errors in the statistical analyses in about 50% of the papers using such analyses [4, 12]. Various kinds of errors have added up to this alarming number, but typically the errors were simple, for instance due to misconceptions regarding the basic principles of the testing of hypotheses. Another common example of the misuse of statistical procedures is uncritical application of methods based on assumptions of Gaussian (‘normal’) distributions [see e.g. 5]. We will focus on this latter problem, while noting that besides being important in itself, it is an example of the large class of problems arising from application of standard methods without carefully checking their adequacy in each specific situation.