Sample Size for Dissertation Research: A Practical Calculator Guide
By Dr. David C., PhD in Biostatistics. Reviewed by Dr. Patricia N.. 2026-05-16.
Get expert help with this chapter | University guides | Dissertation writing services
Key takeaways
- Always run a formal power analysis for quantitative work; cite the effect size source.
- For qualitative work, document saturation with a code-per-interview graph.
- Defend small samples with recruitment numbers, sensitivity analyses, and acknowledged limitations.
- Mixed methods justify each strand separately.
- G*Power is free and standard; R's pwr package is equivalent.
Why sample size matters at the defense
The single most common methodology challenge at US dissertation defenses is sample size justification. A defensible answer cites a formal calculation, names the effect size assumption, and acknowledges the limitations.
An indefensible answer cites convenience ('this is who I could recruit') without showing power analysis or saturation logic.
Quantitative: power analysis
Use G*Power (free) or pwr in R. Inputs needed: expected effect size, alpha (typically .05), desired power (typically .80), number of predictors or groups.
Defaults for medium effects: t-test ~64 per group, correlation ~85 total, multiple regression with 5 predictors ~92 total, factorial ANOVA 2x2 ~128 total, SEM ~200 minimum.
Justify your effect size assumption with prior published research in the area, not a guess.
Qualitative: saturation logic
Saturation is when new data stops producing new themes or codes. Document the point at which you reached it.
Common guidelines: phenomenology 6 to 15, grounded theory 20 to 30, case study 1 to 5 cases, ethnography 30 to 50 hours of observation. These are starting points, not rules.
Defend your sample by showing the saturation curve: a graph of new codes per interview. The line should flatten before you stop collecting.
Mixed methods
Justify each strand separately. The quantitative strand uses power analysis. The qualitative strand uses saturation.
If the strands have different sample sizes, explain why in the methodology chapter.
Practical calculator examples
Multiple regression, 6 predictors, medium effect (f-squared = .15), alpha .05, power .80: minimum 98 cases.
Logistic regression, balanced classes, OR = 1.5: minimum 250 cases for stable estimates.
Independent samples t-test, large effect (d = .80): minimum 26 per group.
Structural equation model, 3 latent variables with 4 indicators each: minimum 200 cases (10 per estimated parameter).
How to defend a small sample
If recruitment was hard, document the recruitment effort with numbers (e.g. 'invitations sent to 450 potential participants, 38 enrolled').
Run sensitivity analyses or a Bayesian analysis that does not assume large-N asymptotics.
Acknowledge the limitation in Chapter 5 and frame findings as exploratory or preliminary.
Cite a similar published study with a comparable sample size as a precedent.
How DissertlyPro handles sample size
Every quantitative engagement starts with a G*Power or pwr-based calculation, documented in a methodology memo for your committee. Qualitative engagements include a saturation tracking template you populate during data collection.
Need a statistician on your committee timeline? Book a free 20-minute call with a PhD biostatistician on the DissertlyPro team.
FAQ
What if my effect size is unknown?
Use Cohen's conventions (small, medium, large) as defaults and justify the choice. Run a sensitivity analysis showing the detectable effect at your achieved sample.
Can I add cases after analysis if power is low?
Yes, but acknowledge it; do not change your sample size mid-analysis without disclosure.
Does saturation apply to surveys?
No, saturation is a qualitative concept. Surveys use power analysis.
What about pilot study sample size?
Typically 10 to 30 cases. Use the pilot to estimate the effect size for your main study power calculation.