ACADEMIC SUPPORT · UAE & GCC
How to Justify Dissertation Sample Size: Power, Precision and Recruitment
Choose a defensible target for your actual research question, then explain how you will reach it.

A dissertation sample size should follow the study aim, design and planned analysis. If you are estimating a population proportion, plan the precision of the estimate; if you are testing a difference, plan for the smallest meaningful effect and adequate power; if you are interviewing participants, explain how the sample will provide relevant information. Then account for recruitment, nonresponse and exclusions. There is no universal “30 participants” rule or single calculator that works for every study.
A good justification makes its assumptions visible. It names the target population, the primary outcome, the sampling method, the calculation or qualitative rationale, and the number of usable cases required. It also shows how many people may need to be approached to reach that number. Your supervisor, ethics committee and programme handbook may require a particular method or reporting format; check those before collecting data.
This guide is for researchers preparing a proposal or methods chapter across the UAE and GCC. Its examples are fictional teaching cases, not a substitute for a statistician’s design review or your institution’s requirements.
Start with the question, not a sample-size calculator
“How many responses do I need?” sounds simple but hides several different questions. A cross-sectional survey estimating a percentage is different from a two-group comparison, a regression model or a qualitative interview study. The sample-size guide for health researchers explains that analysis, precision, power, confidence and the smallest meaningful effect all shape the target. A more recent methodology guide across study designs likewise cautions against a universal formula.
| Primary task | Planning question | Evidence to record |
|---|---|---|
| Estimate a proportion or mean | How precise should the estimate be? | Expected variability, confidence level, desired interval width and sampling design. |
| Compare groups or test an effect | What is the smallest difference worth detecting? | Primary outcome, effect assumption, power, significance level, allocation and planned test. |
| Model several predictors | How many usable events or cases can support the specified model? | Outcome frequency, candidate predictors, validation plan and specialist method guidance. |
| Qualitative interviews | Will the participants provide enough relevant information for the analytic aim? | Sampling logic, participant variation, interview scope and an explicit stopping/review plan. |
Choose the row that matches your primary research question. Secondary analyses may need different sample considerations; do not claim one calculation guarantees power for every hypothesis.
Write the assumptions before calculating
For a quantitative study, state the population and inclusion criteria, design, primary outcome and analysis. Estimate expected variability or effect from prior comparable evidence where possible. Specify whether participants are independently sampled, clustered by class or organisation, or selected by convenience. The same nominal number can imply different precision under those designs. A design effect, multiple groups, missing outcomes or an imbalanced allocation can materially change the required number.
Outcome first
Define the one estimate or comparison that drives the design. “Student attitudes” is too broad; “proportion of eligible students reporting X” is calculable.
Assumptions visible
Record the effect, variability, power or precision target and where each assumption came from. Label judgement where no reliable pilot estimate exists.
Feasibility tested
Document who can be approached, expected response, exclusions and permissions. A mathematically ideal target is not a recruitment plan.
The Reference Guide on Survey Research explains why sampling design and absolute sample size matter for precision. If your sampling is nonprobability-based, a conventional margin of error should not be presented as though respondents were randomly selected from the whole GCC population.
A worked survey example: separate usable responses from invitations
Imagine a fictional UAE university project estimating the share of students who use a particular library service. Its target population and sampling frame are defined, and the team plans a simple random sample. If it has no reliable prior proportion, it uses 0.5 as a conservative planning value. For a 95% confidence level and approximately ±5 percentage-point margin under that simple design, the familiar large-population calculation is 1.96² × 0.5 × 0.5 ÷ 0.05² ≈ 384.16, rounded up to 385 usable responses. This is an illustration, not a universal dissertation requirement.
Now suppose the team expects only 60% of invited students to return usable questionnaires. It would need to approach about 385 ÷ 0.60 = 641.7, rounded up to 642 eligible people under that assumption. If the expected response is lower or some returned forms lack the primary item, the invitation target must rise. If students are sampled by class, clustering may require a larger usable sample; if the frame is small, the design may require a finite-population adjustment. A statistician should check the final calculation and the institution’s ethics limits.
For comparisons, use the effect you actually care about
A two-group study might ask whether one teaching approach changes a defined score relative to another. Its planning calculation needs a meaningful difference, expected spread, significance level, power and group allocation. Choosing a very large effect because it produces a convenient small sample can leave the study unable to detect a practically important smaller difference. Borrowing an effect from an unrelated population can be equally misleading.
For regression, machine learning or instrument validation, avoid quoting a fixed “participants per variable” or “responses per questionnaire item” rule as sufficient proof. Outcome frequency, model complexity, validation and design-specific guidance matter. The correct analysis and specialist review may change the target materially. The practical sample-size guide and cross-design methodology guide both support design-led planning rather than one formula for every question.
Qualitative studies need a different justification
If you are conducting interviews to understand an experience, an online quantitative power calculator is not the right proof. Explain why the selected participants have direct knowledge of the phenomenon, which perspectives must be represented, how broad the interview question is, and how you will assess whether additional interviews add material insight. A small focused sample may be credible when participants are highly specific and interviews are rich; a diverse multi-country study may need a broader plan. Institutional expectations and the analytic approach still control the final design.
For a mixed-methods dissertation, justify the quantitative and qualitative strands separately, then show how they connect. Do not use the quantitative sample target as a substitute for qualitative sampling logic. The study-design methodology literature discusses this distinction explicitly.
Turn the rationale into a transparent methods paragraph
A reader should be able to reconstruct your decision without guessing at a calculator’s settings. A useful paragraph names the design and primary outcome, the source and value of each assumption, the method used to derive the usable target, and the allowance for nonresponse. It states the achieved sample later, along with deviations from the plan. If the target was constrained by access, say so and discuss how that affects precision or interpretation. Do not use retrospective “observed power” to disguise a recruitment shortfall; the STROBE explanation and elaboration warns against post hoc justifications of study size in observational reporting.
| Weak wording | More transparent wording |
|---|---|
| “A sample of 100 was adequate because 100 is standard.” | “The primary outcome was X; the planning method used Y assumption from Z source. It required N usable cases. We approached M eligible people based on an expected response of R%; the final usable sample and limitations are reported below.” |
| “The online calculator said 385.” | “For a defined simple random survey frame, 95% confidence, ±5 percentage-point target and conservative 0.5 proportion gave 385 usable responses before nonresponse adjustment.” |
These are structural examples, not text to copy without adapting to your real study. Your supervisor should approve the analysis and the assumptions before recruitment.
Final sample-size checklist
- State the primary research question and outcome precisely.
- Match the calculation or rationale to the study design and analysis.
- Source assumptions; distinguish evidence from judgement.
- Account for sampling design, clustering, exclusions and missing responses where relevant.
- Convert the usable-case target into a realistic contact and recruitment plan.
- Confirm permissions, ethics and programme-specific expectations.
- Record the planned and achieved numbers separately, and explain any difference.
If you are still choosing a sampling approach, read Labeeb’s convenience-versus-stratified sampling guide. Its choice affects what you can legitimately infer from the eventual sample.
Sources: Sample size determination: practical guide; methodology across study designs; Reference Guide on Survey Research; STROBE explanation. Check current programme guidance and the specific method literature for your design.
Make the sampling decision defensible
If your sampling rationale is unclear, Labeeb can help you scope an ethical methods review, align the written explanation with your approved analysis plan and check that the recruitment assumptions are transparent. Your supervisor and institution retain authority over the study design, calculations and ethics approval. Share your research question, draft protocol and any prescribed guidance before we define the support needed.
Research Question vs Hypothesis: Choose and Align Your Dissertation
Dissertation Data Analysis Plan: Map Questions to Variables and Tests