No Estimation without Inference

Abstract

The International Society of Physiotherapy Journal Editors (ISPJE) recently published an editorial warning that many of their journals would soon prohibit the use of null hypothesis tests and instead require authors to interpret 95% confidence intervals relative to clinically important values. Although I encourage the reporting of confidence intervals and the discussing of uncertainty in the context of a research question, the ISPJE’s proposed ban is illogical and there are several instances of flawed statistical reasoning in the editorial. In brief, the editorial: (1) fails to adequately grapple with the inherent connection between hypothesis testing and estimation, (2) presents several misleading arguments about the perceived flaws of hypothesis tests, and (3) presents an alternative to hypothesis testing that is, in itself, a form of hypothesis test – the minimal effects test – albeit done informally. If the editorials’ arguments are taken at face value, then that will lower the statistical literacy in our field and readers will have a flawed understanding of p-values. Further, if the editorials’ proposed ban is put into practice, I fear that could decrease the scientific integrity of our research as it removes quantitative benchmarks in favor of a more subjective interpretation of confidence intervals. Ultimately, I think that many of the ISPJE’s concerns that led to the editorial are valid, but I think those problems are the result of questionable research practices stemming from poor methodological training for authors, reviewers, and editors. These problems will only be fixed through better and continuing education, not the banning of statistically valid methods.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 5 citations exist upstream.