Title: “Divided by a common language”: Rebooting statistical basics for biomedical research
Abstract: Poor methodological quality of much biomedical research is increasingly recognized as a major obstacle to translation and reproducibility. The stakes are high: substandard research represents a huge financial and ethical liability, wastes money and research animals, and jeopardises patient welfare. Surveys of published literature indicate biomedical investigators still cling to statistical methods and tools that became outdated in the 1920’s; I present data to show that investigators still do not conduct experiments according to Fisher’s basic principles of experimental design. Research quality is further compromised by profound and widespread misunderstanding and misapplication of the most basic statistical concepts (such as replication and randomisation), and a failure to appreciate the role of simple arithmetic. I use simulations to illustrate the statistical consequences for error estimation, F-distributions and p-values. Most identified problems may seem elementary, but to date have been impossible to solve; investigator culture shows little signs of movement. To put needed changes into effect, we as statisticians urgently need to change our philosophy, language, and outreach strategies.