Calculate your dissertation sample size before collecting data. Use a power analysis for hypothesis testing or a precision calculation to estimate proportions. Your research questions and planned analysis determine which method fits.
There isn’t one participant count that suits every dissertation. A defensible target accounts for meaningful effects, uncertainty and incomplete responses, rather than copying another student’s recruitment figure.
For quantitative research, distinguish the number to analyse from the number to recruit when making the sample size calculation.
Key Takeaways
- Base your calculation on the primary research question, planned statistical test and justified assumptions.
- Allow for dropout, missing data and non-response after calculating the required analysable sample.
- Report your calculation clearly, and remember that increasing sample size doesn’t remove recruitment bias.
Choose the Inputs for Your Dissertation Sample Size
Connect Power, Effect Size and Variability
Statistical power is the probability of detecting a specified effect when that effect exists. Common planning targets are 80% or 90%. The significance level, usually α = 0.05, controls the false-positive probability under the test’s assumptions.
For a specified study design, sample size, effect size, significance level and power are linked. Fixing three determines the fourth. Variability also matters: detecting a given difference becomes harder when measurements vary more.
For standardised mean differences, Cohen’s benchmarks are d = 0.2, 0.5 and 0.8, conventionally described as small, medium and large. However, a meaningful difference in your subject should guide the choice.
Match the Calculation to the Design
Use power analysis when testing differences or relationships. For a descriptive survey estimating prevalence, the desired confidence interval width may matter more.
Identify your primary outcome, comparison groups and unit of analysis first. Repeated measurements from one participant aren’t independent participants, whilst clustered recruitment may require a design-effect adjustment.
Our guide to dissertation sampling methods explains how participant selection affects the conclusions your sample can support.

Run a Power Calculation in G*Power
Work Through an Independent-Samples t-Test
The free G*Power statistical power software supports many common dissertation analyses. For two independent groups with equal allocation:
- Select “t tests”, then “Means: Difference between two independent means (two groups)”.
- Choose “A priori” to calculate the required sample size.
- Set two tails, effect size d = 0.5, α = 0.05, target statistical power of 80% and allocation ratio = 1.
- Select “Calculate” and record the inputs and output.
These settings require 128 participants, with 64 per group. Changing only the effect size to d = 0.2 increases the requirement to 788 participants, or 394 per group.
This difference explains why detecting subtle effects can exceed a dissertation’s recruitment capacity. Don’t choose a larger effect solely to obtain a convenient target.
Calculate a Multiple-Regression Example
For an overall multiple-regression model with five predictors, select “F tests”, then “Linear multiple regression: Fixed model, R² deviation from zero”.
Choose an a priori calculation and enter f² = 0.15, α = 0.05, power = 0.80 and five predictors. The required total is 92 complete observations.
Here, f² is a regression effect-size measure, not Cohen’s d. This regression analysis tests the overall model; detecting one predictor’s additional contribution needs a different calculation. Count model parameters correctly, including dummy variables for categorical predictors.
Calculate a Survey Sample Using Cochran’s Formula
Estimate a Population Proportion
For a simple random sample, the sample size calculation for a population proportion uses Cochran’s formula:
n₀ = Z²pq / e²
Z is the confidence-level z-score, p is the expected proportion, q = 1 − p, and e is the absolute margin of error. At a 95% confidence level, Z = 1.96.
If p is unknown, use 0.5. This maximises p(1 − p), producing a conservative requirement, including when the true proportion lies between 10% and 90%.
With a five-percentage-point margin of error:
n₀ = 1.96² × 0.5 × 0.5 / 0.05² = 384.16
Round up to 385 completed responses. Penn State’s confidence interval and sample-size guidance explains the underlying proportion calculation. OpenEpi also provides free sample-size tools.
Apply the Finite Population Correction
When sampling without replacement from a small, known population, apply:
n = n₀ / [1 + (n₀ − 1) / N]
For N = 1,000 eligible students, the calculation becomes:
384.16 / [1 + 383.16 / 1,000] = 277.74
You therefore need 278 completed responses, before adjusting for non-response.
These calculations assume an appropriate probability sampling design. A social-media convenience sample doesn’t acquire a defensible population margin of error merely because it reaches 385 responses.
Increase the Target for Attrition and Non-Response
Calculate the analysable sample first, then adjust recruitment:
Recruitment target = required sample / (1 − expected loss rate)
For 200 analysable participants and 10% expected attrition:
200 / 0.90 = 222.2
Round up to 223 recruits. Clinical planning sometimes uses an allowance around 10% for dropout or non-compliance, but previous comparable studies provide a better basis than a default percentage.
For a survey expecting 15% burnout prevalence, a 95% confidence level and a five-percentage-point margin of error:
1.96² × 0.15 × 0.85 / 0.05² = 195.9
Round up to 196 completed responses. With 20% non-response, the expected response rate is 80%, so recruit or invite 196 / 0.80 = 245, assuming everyone approached is eligible.
Distinguish initial non-response from later dropout. If both apply, account for both stages rather than treating them as interchangeable.
Write a Sample-Size Justification Examiners Can Assess
A clear sample size justification should report the primary analysis, software and version, effect-size rationale, significance level, desired power and allocation assumptions. State the analysable target and any recruitment adjustment.
For the worked t-test, proposal wording could read:
An a priori power analysis in G*Power indicated that 128 participants, allocated equally between two independent groups, would provide 80% power to detect d = 0.5 using a two-tailed t-test at α = 0.05. Allowing for 15% attrition, the recruitment target is 151 participants.
The effect-size assumption still needs supporting evidence. Relevant previous studies or a substantively meaningful difference are stronger grounds than choosing “medium” without explanation.
A dissertation pilot study can assess feasibility and measurement procedures, although effect estimates from small pilots are unstable.
Also explain sampling restrictions and missing-data handling. Our dissertation methodology examples show how to connect procedures with their justification. In completed research, report what happened alongside what you planned.
Choose Defensible Analyses When Your Sample Is Small
Match Alternatives to the Data
A small sample size doesn’t automatically require nonparametric tests. Your question, measurement scale, study design and assumptions determine suitability.
The Mann-Whitney U test can compare independent groups, Wilcoxon signed-rank addresses paired observations, and Kruskal-Wallis compares independent groups beyond two. Spearman correlation assesses monotonic associations. For sparse contingency tables, Fisher’s exact test may suit a two-by-two comparison.
However, these methods don’t restore missing information or automatically test the same hypotheses as parametric procedures. Mann-Whitney U, for example, doesn’t universally test differences between medians. Our guide to statistical tests for dissertations explains these distinctions.

Avoid Fragile Models and Misleading Interpretations
Regression generally requires more information as predictors increase. The commonly cited rule of thumb of ten observations per variable is contested and doesn’t establish adequacy.
Logistic regression presents additional problems when outcomes are sparse. A dataset containing 45 participants with the outcome and only three without provides little information about the smaller category. Perfect separation, where a predictor completely distinguishes outcomes, can prevent ordinary maximum-likelihood estimates from remaining finite.
If recruitment falls short, report estimates, confidence intervals and p-values, and state the resulting limitations. Use descriptive statistics to show distributions when inferential analysis is limited. Consider a sensitivity analysis showing the effects your achieved sample could detect.
Avoid calculating observed power from a non-significant result, as CONSORT reporting guidance cautions. A non-significant finding doesn’t establish that no relationship exists.
Frequently Asked Questions About Dissertation Sample Size
Is There an Ideal Sample Size, and Can I Justify 30 Participants?
There is no universal ideal. Thirty participants may support a feasibility study, descriptive exploration or a design capable of detecting a sufficiently large effect. However, 30 isn’t a general statistical threshold.
Justify the number against the study’s purpose, expected precision or detectable effect. If access determines the available sample, state that constraint and limit your claims accordingly. Don’t describe a restricted sample as adequately powered without a calculation.
For an existing dataset, calculate sensitivity or expected precision before choosing the final analysis. Keep that assessment separate from observed power calculated using the results.
How Many Interviews Does a Qualitative Dissertation Need?
Power formulas used in quantitative research don’t determine adequacy in qualitative research. Depending on the approach, use information power or a clearly defined saturation argument.
Hennink and Kaiser’s 2022 systematic review, cited in UNC’s qualitative writing resources, found saturation often occurred within about nine to 17 interviews in relatively homogeneous studies with narrow objectives.
That range isn’t a recruitment rule. Participant diversity, interview depth and analytical approach change what is needed. Where saturation fits your methodology, analyse during recruitment and document when additional interviews stop contributing relevant information. Explain how you assessed it.
Build Your Target Around the Evidence You Need
A defensible dissertation sample size begins with the primary question and a calculation that matches your analysis. Record your assumptions, adjust for expected losses and distinguish recruitment numbers from usable observations.
If a small sample size is unavoidable, narrow the question or acknowledge reduced sensitivity. A clearly bounded study is more credible than an unsupported claim that the sample is sufficient.

