Statistical methods in epidemiology: I. Statistical errors in hypothesis testing

Abstract

Purpose: Although scientific journal editors are making use of statisticians in the review process, the quality of statistical reporting in many journals remains poor. In many cases the problem for the scientist would appear to be a lack of understanding of basic statistics. The focus of the scientist is on showing 'p < 0.05', when what is actually required is a statement about effect size and interval estimation. The aim of this paper is to show the inadequacy of reporting of results using p-values alone. This paper is the first in a series detailing common statistical methods, with a view to aiding potential authors in their statistical presentation of data. Method: A review of the basic hypothesis test, using examples from the author's own teaching experiences. Results: Type I and type II errors are defined; the problem of multiple comparisons is highlighted; interval estimation is introduced. Conclusions: The case for considering the p-value as an error probability is made which suggests ways of improving statistical presentation and thus expediting the statistical review process.