Top 10 Statistical Mistakes in Dissertations (And How to Fix Them)
By Dr. David C., PhD in Biostatistics. Reviewed by Dr. Robert K.. 2026-05-14.
Get expert help with this chapter | University guides | Dissertation writing services
Key takeaways
- Power analysis, effect sizes, and assumption checks are the three most common gaps.
- Use APA 7 p-value formatting: 'p < .001', never 'p = .000'.
- Match the analysis to the data structure; pseudo-replication fails defenses.
- Disclose and handle missing data explicitly; never silently delete.
- Report intercoder reliability for qualitative work.
Why these mistakes matter
Every committee statistician has a short list of errors they look for first. Find these in your draft now, fix them, and you remove the most common reasons for major revisions.
1. No power analysis
The problem: no justification for sample size beyond 'this is what I recruited'.
The fix: run G*Power or pwr in R. Document inputs (effect size, alpha, power) and outputs in the methodology chapter.
2. Misreporting p-values
The problem: 'p = .000' or 'p = ns' instead of APA 7 format.
The fix: 'p < .001' when below threshold; exact value to three decimals otherwise.
3. Ignoring assumption checks
The problem: running parametric tests without checking normality, homogeneity, or sphericity.
The fix: report Shapiro-Wilk or visual inspection for normality, Levene's for variance, Mauchly's for sphericity, with corrections applied when violated.
4. Misinterpreting non-significance
The problem: 'no effect was found' instead of 'no significant effect was detected'.
The fix: language matters. Non-significance does not equal absence of effect; consider confidence intervals and equivalence testing.
5. Reporting significance without effect size
The problem: F-statistic and p-value without partial eta squared or Cohen's d.
The fix: APA 7 requires effect size. Add it to every test.
6. Confusing correlation and causation
The problem: cross-sectional correlational data described with causal language.
The fix: 'is associated with' rather than 'causes'. Reserve causal language for experimental designs.
7. Pseudo-replication
The problem: treating repeated measurements on the same participants as independent observations.
The fix: use multilevel modeling, repeated-measures ANOVA, or mixed models. Never independent-samples tests on clustered data.
8. Over-interpreting subgroup analyses
The problem: running 20 subgroup tests and reporting only the significant ones.
The fix: pre-register subgroup analyses or apply Bonferroni or Holm correction. Acknowledge exploratory tests as such.
9. Missing data handled poorly
The problem: listwise deletion without disclosing it, or worse, mean-substitution.
The fix: report missingness, use multiple imputation or full information maximum likelihood when appropriate, defend the choice.
10. Inadequate reporting of qualitative coding
The problem: themes presented without showing the analytic process.
The fix: report intercoder reliability (Cohen's kappa or Krippendorff's alpha), describe the coding scheme, and provide an audit trail.
How DissertlyPro catches these before the defense
Every DissertlyPro statistics engagement includes a defense-prep pass that checks for all 10 mistakes above. We supply the syntax, the corrections, and a written methodology memo defending each choice to your committee.
Need a statistician on your committee timeline? Book a free 20-minute call with a PhD biostatistician on the DissertlyPro team.
FAQ
Which mistake is most common?
Missing or weak power analysis. Almost every dissertation we review needs strengthening here.
Do qualitative dissertations have statistical mistakes?
Different kind: inadequate intercoder reliability, weak saturation evidence, and confusion of themes with categories.
How do I know which assumption checks to report?
Match to the test: t-test (normality, variance), ANOVA (normality, variance, sphericity for repeated measures), regression (linearity, normality of residuals, homoscedasticity, independence).
Can a statistician fix these in my draft?
Yes. A PhD statistician can rerun analyses, apply corrections, rewrite the results chapter, and produce a methodology memo for your committee.