Description
AIMOS attracts experts in meta-research, open science, and evidence-based policymaking, offering valuable networking opportunities with researchers and policymakers who share an interest in funding impact assessment. Engaging with these experts will enhance our evaluation framework, ensuring it remains rigorous, scalable, and aligned with best practices. The conference will also allow us to receive feedback on our methodology from interdisciplinary perspectives, strengthening our research’s applicability across different funding schemes.Additional information
Lightning Talk was presented at AIMOS 2025:Background
Statistical errors can lead to misleading outcomes, harm patients, and erode trust in research. This review categorises and quantifies such errors in health studies.
Methods
We included studies analysing statistical errors in health journal articles. Three databases were searched, screened by three authors. One author extracted data, another checked it. Generalisability was assessed using a modified RoB-PrevMV tool.
Results
From 3,421 articles, 53 were included: 38 on various errors and 15 on specific ones. The first group listed 442 Data Analysis errors, and the second 31, based on Mansournia’s CHAMP Statement. Thirty-five statistical errors were cited in four or more articles, and eight errors were reported by at least 10 studies. “General incorrect statistical analysis” is reported most frequently (40 times), with prevalence ranging from 7%-96%, with median 22%. Of errors reported in >10 articles, “Failure to state the number of tails” has the highest median (67%, range 5–98%), and “Reporting paired vs independent tests” has the lowest (8%, range 0–58%).
Conclusion
Statistical errors are highly prevalent in the health research literature and are highly heterogeneous across the included studies. There is a continued need for improvements in research methodology, education, and editorial oversight.
| Period | 19 Nov 2025 → 21 Nov 2025 |
|---|---|
| Event type | Conference |