Pitfalls and misuse in data analysis

Data analysis is one of the cornerstones of biomedical research and evidence-based medicine. However, the conclusions it produces and the body of health applications derived from it are undermined by a variety of statistical pitfalls, common errors, and tolerated methodological malpractice. The persistence of these pitfalls in the literature alters the nature of interpretations and acts as a brake on scientific discoveries and consensus. In the context of a massive and growing flow of publications and non-expertised results (pre-publications, communications on social networks), methodological rigor and solid statistical training are more than ever the best bulwarks against a crisis of reproducibility and confidence. In this article, we draw on recent methodological literature to review the main methodological pitfalls, be they well-known ones such as those relating to the significance or the multiplicity of comparisons, or less familiar ones such as dichotomisation, variable selection, or problems of spatial and temporal auto-correlation. This approach is part of a long series of refinements and reminders initiated several decades ago by the methodological community to maintain a level of control and criticism that is essential to the reliability of biomedical research.