From Inference to Ritual: Why Frequentist Statistics Became Hard to Understand and Teach in Medicine

Frequentist statistical tools still structure a large part of biomedical research, regulatory evaluation, and medical teaching, even though their correct interpretation remains uncertain among students, clinicians, and readers of the literature. This misunderstanding is usually attributed to pedagogical shortcomings or individual cognitive errors. This article examines the structural reasons for this persistent misunderstanding. It argues that it arises from a combination of historical, methodological, semantic, and institutional factors rather than from a direct lack of technical training.

The paper uses a selective narrative and historical analysis of the Bayes-Fisher-Neyman trajectory, of the language used to describe major frequentist tools, of the critiques of their ritualized use, and of their institutionalization in editorial, regulatory, and curricular settings. Its purpose is synthetic and argumentative, with a focus on medical education.

The article argues that, in Fisher's framework, the p value was not an isolated signal but one element within a dense experimental architecture based on study design, sufficient statistics, and exhaustive extraction of information. The later shift toward a procedural and decision-oriented meaning strengthened its widespread use as a simple decision tool but made its frequentist meaning less natural. This opacity is further reinforced by ambiguous terminology, by the separation between statistics and methodology in teaching, and by ritualization through editors, agencies, and curricula.

Misunderstanding of frequentism in biomedicine should be seen as a structural problem of intelligibility. A realistic educational response requires reconnecting calculation, study design, history, epistemology, and language, so that teaching moves away from ritual use of tools and toward a more robust understanding of their meaning and limits.