Common Statistical Mistakes in Medical Dissertations
The statistical errors examiners most often find in MD, MS, DNB and MDS dissertations, with simple ways to avoid each one in your analysis and write-up.
Examiners read many dissertations, and the same statistical problems appear again and again. Most are easy to avoid once you know them. Here are the ones to check for before you submit.
1. Sample size with no justification
A number without a formula, inputs and a reference raises immediate questions. See our sample size guide.
2. The wrong test for the data
Common examples: a t-test for skewed data in small groups, an independent test for before–after data, or chi-square when expected counts are very small (use Fisher's exact test). Decide the test from the data type, design and distribution.
3. Many tests without a plan
Running dozens of comparisons and reporting the significant ones inflates false positives. Base your analysis on the objectives, and adjust for multiple comparisons where needed (for example, post-hoc tests after ANOVA).
4. Only p-values, no effect sizes
"p < 0.05" does not tell the reader how big the difference is. Report the mean difference, odds ratio or relative risk, with a 95% confidence interval and the exact p-value.
5. Mean (SD) for skewed data
For variables such as length of stay or income, report the median and interquartile range. Check distributions with histograms.
6. Percentages without denominators
Write "48 of 120 (40.0%)" rather than "40%". When percentages are calculated from different totals because of missing data, show the denominators.
7. "No difference" from a non-significant result
A non-significant p-value is not proof of no effect, especially in small studies. Report the confidence interval and discuss whether the study had enough power.
8. Correlation treated as causation
An association in an observational study may be explained by confounding or reverse causation. Use careful language such as "was associated with".
9. Confounding ignored
If groups differ in age, sex or severity, a simple comparison can mislead. Consider stratified analysis or regression to adjust for important confounders.
10. Results that do not follow the objectives
Organise the results chapter by objective, in the same order as in the synopsis. Readers should find the answer to each objective easily.
11. Tables and graphs that do not stand alone
Every table and figure needs a clear title, units, the number of participants and the test used. Avoid 3D charts and pie charts with many slices.
12. Conclusions beyond the data
A hospital-based study of 100 patients cannot establish what is true for the whole country. Keep conclusions within your sample and design.
Quick self-check
| Question | Yes / No |
|---|---|
| Is the sample size justified with a formula and reference? | |
| Does each objective have a matching analysis? | |
| Are effect sizes and 95% CIs reported with p-values? | |
| Are skewed variables summarised with median (IQR)? | |
| Are denominators shown for percentages? | |
| Is the language about associations careful and accurate? |
Key takeaways
- Plan the analysis from the objectives, before seeing the data.
- Match tests to data types and designs.
- Report effect sizes, confidence intervals and exact p-values.
- Write conclusions that your design and sample can support.
References
- Lang TA, Altman DG. Basic statistical reporting for articles published in biomedical journals: the "Statistical Analyses and Methods in the Published Literature" or SAMPL guidelines. In: Smart P, Maisonneuve H, Polderman A, eds. Science Editors' Handbook. European Association of Science Editors; 2013.
- Altman DG, Bland JM. Absence of evidence is not evidence of absence. BMJ. 1995;311(7003):485.
- Wasserstein RL, Lazar NA. The ASA statement on p-values: context, process, and purpose. Am Stat. 2016;70(2):129–133. doi:10.1080/00031305.2016.1154108
- Bland JM, Altman DG. Multiple significance tests: the Bonferroni method. BMJ. 1995;310(6973):170.
This article is for education. Always follow the requirements of your university, guide, ethics committee and target journal.