Sabitlenmiş Tweet
I am delighted to share the findings of our recent study, "Some misinterpretations of inferential statistics in dental public health literature," conducted in collaboration with Lubna Al-Nasser, Walid Al-Soneidar Al-Soneidar, and @Mohamma70696197, an international expert in epidemiologic methods, statistics, and causal inference.
To build on earlier calls on proper utilization and reporting (American Statistical Association’s 2016 statement: lnkd.in/evhpCUjf), we examined how p-values and confidence intervals (CIs) were reported and interpreted in dental public health journals. We found several concerning patterns:
1. P-values were reported much more frequently than CIs (not wrong by itself).
2. Dichotomization (p < 0.05 vs p ≥ 0.05) was highly prevalent.
3. Methods sections often explicitly stated that p < 0.05 indicates statistically significant findings.
4. Non-significant results (p≥ 0.05 or CIs including the null) were often misinterpreted as “no effect.”
5. CIs were frequently used as a surrogate significance test, ignoring their informative range.
6. No meaningful improvement in reporting was observed between publications closer and further from the American Statistical Association’s 2016 statement.
7. Reporting differed between journals.
Key Messages:
1. Table 1 and Appendix Table 9 provide readers with correct and incorrect interpretations of p-values/CIs (synthesized from the relevant literature) and applied to a hypothetical research scenario.
2. We echo the calls to abandon reporting p-value as inequality or dichotomizing p-values/CIs.
3. P-values should be interpreted continuously (or interpreted using surprisal).
4. CIs interpretation should use the full range (use compatibility).
Full article: lnkd.in/ejxGB7se
#Statistics #Epidemiology #ResearchMethods #PublicHealth #DentalPublicHealth #EvidenceBasedPractice #AcademicIntegrity
English
