Hypothesis-generating and confirmatory studies, Bonferroni correction, and pre-specification of trial endpoints

摘要

A p-value presents the outcome of a statistically tested nullhypothesis. It indicates how incompatible observed data arewith a statistical model defned by a null hypothesis. Thishypothesis can, for example, be that 2 parameters have identical values, or that they differ by a specifed amount. A lowp-value shows that it is unlikely (a high p-value that it is notunlikely) that the observed data are consistent with the nullhypothesis. Many null hypotheses are tested in order to generate study hypotheses for further research, others to confrm an already established study hypothesis. The differencebetween generating and confrming a hypothesis is crucial forthe interpretation of the results. Presenting an outcome froma hypothesis-generating study as if it had been produced in aconfrmatory study is misleading and represents methodological ignorance or scientifc misconduct.