The first context was when reviewing a paper proposing methods to investigate the quality of published clinical trials. It was suggested as an indicator of valid randomisation in, and valid conduct of, a clinical trial. A reviewer could look at the distribution of the P-value in comparisons of baseline data between the randomised groups. All the null hypotheses should be true, so this distribution should be Uniform if participants were randomised fairly. Many triallists would object that such tests are meaningless because if all null hypotheses are known to be true then any positive significance tests must be type 1 errors. From this viewpoint, the presence of such tests is an indicator of bad reporting and low quality. However, reviewers can do the tests themselves, given published means and standard deviations or counts. A further objection is that such tests within a study are not independent. For example, triallists might report variables such as height and weight, or diastolic and systolic blood pressure. These variables would be correlated. Would the tests have to be independent for this idea to work?
