——> Follow us on TWITTER

Kill or over-kill of p-values

Readers of medical journals know that only few papers are published without one or more p-values. Some of these readers believe they understand the meaning of these p-values. They are usually wrong as are most of the editors or peer-reviewers. The better educated epidemiologist knows that p-values are complicated to interpret. Very few actually know how they can be interpreted and under which conditions these interpretations apply.

Why then provide p-values in scientific papers? Many would say “let’s drop the statistics altogether”. Others would argue the p-value is a useful statistic and one should educate users of what p-values are and what they are not. Sander Greenland suggests that public health society should endorse a number of statements of what p-values are not and ask societies to endorse the statements (enclosed). We generally agree with Greenland’s statements, but feel that there should be good debate about them before we make a decision about endorsing them. In particular, people who wish to continue using p-values should be able to speak in their defense.

On the other hand, misuse of p-values is not trivial. Some have claimed that misinterpretation of p-values has killed more people that most other types of scientific misconduct. Overlooking valuable treatment or risky exposures because the results were not statistically significant is not the fault of the p-value itself, but is a common problem that arises out of the use of p-values without adequate thought and scrutiny by scientific authors, peer reviewers and editors. Most active researchers know that this problem still exists even now after years of debate. It is, however, positive that some journals, including the IEA’s International Journal of Epidemiology, have taken actions to reduce harm caused by the misinterpretation of p-values. That is all good, but it may not be sufficient.

Endorsing scientific statements is a serious action because truth is a short lived trait. The misinterpretation problem may, however, still remain so serious that it may be justified to take action to reduce the misuse of p-values. What do you think? We welcome comments and opinions that will be posted on our website.

Jorn Olsen, Neal Pearce, Cesar Victora

Share Button