Data presentation and the use of statistical tests in biomedical journals: Can we reach a consensus?

摘要

There is more or less general acceptance of the proposition that theuse and presentation of statistics in biomedical journals is less thanoptimal. The challenge, then, is to improve it. The underlying problemis multifactorial. First, there is a vast array of statistical methodsavailable to researchers, so that knowing when one should use a particularapproach and how to interpret the outcomes is difficult forexperienced researchers, let alone novices. This problem is, in turn,exacerbated in this new age of bioinformatics by the complexities ofthe analysis of large data sets arising from, for example, genetic1 andhigh-throughput genomic ? proteomic studies.2 Second, there is considerableinertia for change among authors, reviewers and editors.We all must be amateur biostatisticians but, unfortunately, few of usare really up to the job and even fewer of us are interested in changingour tried and (more or less) true habits. Third, in some cases thereis still considerable disagreement among biostatisticians as to what isthe correct statistical approach to a particular problem.3 If card-carryingbiostatisticians cannot agree, what hope do humble pharmacologistsand physiologists have?