PhD research guide
How to Choose a Sample Size for PhD Research
How to calculate and justify a sample size for PhD research. Cochran's formula, finite population correction, power analysis with G*Power, and qualitative sample sizes.

Your sample size should be large enough to answer your research questions with the precision or statistical power your analysis needs, and you must justify it with a recognised method. For surveys estimating proportions, Cochran's formula is common; for testing relationships or differences, use a power analysis (for example with G*Power); for structural equation modelling, consult methodological guidance for your model; and for qualitative studies, sample until you reach data saturation, guided by typical ranges for your design.
"My supervisor said 300" or "I could only collect 150" are not justifications on their own. Examiners expect to see how the number was reached.
Step 1: Define your population and sampling frame
- Population: everyone your findings are meant to apply to (for example, nurses in private hospitals in Bhopal).
- Sampling frame: the list or source from which you actually draw participants.
- Sampling technique: probability (simple random, stratified, cluster, systematic) or non-probability (purposive, convenience, snowball, quota).
Step 2: Choose a method that matches your aim
Estimating a proportion or prevalence: Cochran's formula
For a large population:
n₀ = Z² × p × (1 − p) / e²
- Z = 1.96 for 95% confidence
- p = expected proportion (use 0.5 if unknown, which gives the largest sample)
- e = margin of error (for example 0.05)
Example: n₀ = (1.96² × 0.5 × 0.5) / 0.05² = 0.9604 / 0.0025 ≈ 384
Smaller populations: finite population correction
If your population (N) is known and not very large:
n = n₀ / (1 + (n₀ − 1) / N)
Example: for N = 2,000 nurses, n = 384 / (1 + 383 / 2,000) ≈ 322
Testing relationships or differences: power analysis
When you plan t-tests, ANOVA, correlation or regression, base the sample size on:
- Expected effect size (from earlier studies or conventions such as Cohen's small, medium and large)
- Significance level (α), usually 0.05
- Desired power (1 − β), usually 0.80
- Number of groups or predictors
The free program *GPower calculates this. For example, a multiple regression with 5 predictors, a medium effect (f² = 0.15), α = 0.05 and power = 0.80 needs a sample of about 92. Report all the inputs you used.
Structural equation modelling (SEM) and factor analysis
Sample size needs depend on model complexity, the number of indicators per factor, the strength of the loadings and the estimation method. Older rules of thumb (such as 10 cases per indicator, or a minimum of 200) are still cited but are crude. Where possible, base your choice on a power analysis or a simulation approach recommended in the SEM literature, and cite the guidance you followed.
Qualitative studies
Qualitative sample sizes are guided by saturation: the point where new data stop producing new themes. Typical ranges reported in the methodological literature are:
| Design | Common range |
|---|---|
| Phenomenology | about 5 to 25 participants |
| Grounded theory | about 20 to 30 or more interviews |
| Case study | 1 to several cases, each studied in depth |
| Focus groups | several groups of 6 to 10 participants |
Explain how you will judge saturation, and report when you reached it.
Step 3: Adjust for practical realities
- Non-response: divide the required sample by the expected response rate.
- Subgroup analysis: if you will compare groups (for example urban and rural), each group needs an adequate size.
- Stratified sampling: allocate the sample across strata proportionally or as your design requires.
- Data quality: allow for incomplete or unusable responses.
Step 4: Write the justification
A clear justification paragraph in your methodology chapter might read:
The population comprised approximately 2,000 nurses working in private hospitals in Bhopal (source). Using Cochran's (1977) formula with 95% confidence, a 5% margin of error and p = 0.5, the required sample was 384, adjusted to 322 using the finite population correction. Allowing for an expected response rate of 75%, 430 questionnaires were distributed using proportionate stratified random sampling across hospital size categories.
Common mistakes
- Using 384 regardless of the population, aim or analysis
- Sizing the sample for descriptive statistics but then running complex models
- Ignoring non-response and subgroup needs
- Describing a convenience sample as random
- Not reporting the inputs used in a power analysis
Our SPSS and data analysis support and methodology consultations can help you calculate and justify your sample.
Frequently asked questions
Is 384 always the right sample size?
No. 384 comes from Cochran's formula for estimating a proportion at 95% confidence with ±5% margin of error and maximum variability (p = 0.5), for a large population. Different aims, margins, populations or analyses give different numbers.
What about non-response?
Increase the number you approach to allow for expected non-response. For example, if you need 384 usable responses and expect a 75% response rate, approach about 384 ÷ 0.75 ≈ 512 people.
Can I use convenience sampling?
Many PhD studies do, out of necessity, but you must describe it honestly, explain why, and discuss how it limits generalisation. Do not describe a convenience sample as random.