Top 10 statistical pitfalls: a reviewer’s guide to avoiding common errors

Abstract

Milestone articles have highlighted the frequency and types of statistical errors in research,1–5 yet fundamental errors persist across various disciplines. With a background in biostatistics and over 200 articles reviewed for journals such as Heart and Addiction since 2021, two differing and distinct research areas, I (DJG) have identified 10 common statistical mistakes that authors frequently make. Together with two academic colleagues, we present these issues in a concise, direct and accessible way to help researchers avoid them. This article will not repeat the pitfalls documented previously; rather, it reflects independent observations on statistical and presentational issues frequently made by authors across various medical fields. Ever wondered why your manuscript keeps getting rejected? It might be due to these common statistical mistakes. By highlighting these errors, we aim to save authors time with revisions and reviewers time in repeatedly reporting the same issues, ultimately advancing a research world that is transparent, specific and reproducible. Here is the upcoming list (each with their own merit and importance): 1) Incorrect use of causal language: avoid implying causation when only association is demonstrated. 2) Poorly formatted abstracts: ensure abstracts are concise, well-structured and accurately reflect the study. 3) Results in the methods section: keep results strictly within the results section to maintain clarity. 4) Inaccurate/incomplete statistical analysis presentation in the methods: provide detailed and accurate descriptions of statistical methods used. 5) Hypothesis tests for normality: understand when and how to appropriately assess for normality. 6) Absent or insufficient flow diagrams: use flow diagrams to clearly depict study design and participant flow. 7) Regression models issues (three-in-one): differentiate between multivariable and multivariate models, justify confounder inclusion, and ensure model assumptions are met. 8) Using univariable significance for multivariable models: avoid relying on univariable analysis to inform multivariable models. 9) Poor reporting of missing data: transparently report and handle missing data. 10) Insufficient attention …