Direct answer: There is no national minimum. For a quantitative study the number comes from a power analysis based on your test, your expected effect size and your chosen power level. For a qualitative study it comes from a defended sampling logic, not a target. In both cases what is marked is the justification, not the number.
Why does nobody give you a straight number?
Because the honest answer depends on your design, and the people who quote “you need thirty” are repeating a half-remembered rule about when a sampling distribution starts behaving. Thirty is not a South African requirement, a faculty requirement or a statistical requirement. It is a folk number, and an examiner who reads “a sample of 30 was selected as this is the accepted minimum” will ask which authority accepts it.
What replaces the folk number is a sentence with a reason in it. For quantitative work that sentence describes a power calculation. For qualitative work it describes why the people you spoke to were the right people and why you stopped when you did. Both sentences are learnable in an evening, and both are worth more marks than any particular value of N.
What actually determines a quantitative sample size?
Four inputs, and you must decide all four before the calculation means anything.
- The statistical test you will run. A comparison of two groups, a correlation and a multiple regression all have different sample requirements. If you have not settled this, our guide to choosing a statistical test comes first — sample size is downstream of test choice, not the other way round.
- The effect size you expect. The smaller the effect you want to be able to detect, the more people you need. This is the input students find hardest because it feels like guessing; it is not, and the next section deals with it.
- The significance level (alpha). Conventionally 0,05 — the risk you accept of finding an effect that is not there.
- The power you want. Conventionally 0,80 — an 80% chance of detecting the effect if it genuinely exists. Power is the input students omit, and omitting it is what produces studies that find nothing and cannot say whether that means anything.
Feed those four into a power analysis and it returns the minimum N. Reverse the calculation — fix N at what you can realistically collect and solve for power — and you learn something equally useful: whether the study you can actually run is capable of answering the question you are asking.
Where do you get the effect size from?
Three legitimate sources, in descending order of strength. Best: a published study close enough to yours that its reported effect is a reasonable expectation for your context — this is one of the concrete payoffs of having done a proper literature review, because the effect sizes are sitting in the papers you already read. Second best: your own pilot study, with the caveat that small pilots estimate effects very imprecisely. Acceptable and common: a conventional benchmark — the widely used small, medium and large conventions associated with Cohen — with a stated argument for which one is appropriate.
Where students go wrong is assuming a large effect because it produces a comfortably small sample. That is the calculation running backwards from the answer you wanted, and it guarantees an underpowered study. If the literature in your area reports modest effects, plan for a modest effect.
How do you actually run a power analysis?
G*Power, from Heinrich Heine University Düsseldorf, is free, runs on Windows and macOS, and is the tool most South African supervisors will expect to see named. The procedure is the same regardless of test: choose the test family, choose the specific test, select “A priori: compute required sample size”, enter your effect size, alpha and power, and read off the total sample. Screenshot the output and keep it — the input parameters are what you will reproduce in your methodology chapter, and a reader must be able to reconstruct your calculation from your sentence alone.
The methodology sentence looks like this: “An a priori power analysis conducted in G*Power for an independent-samples t-test, assuming a medium effect (d = 0,50), α = 0,05 and power = 0,80, indicated a required total sample of 128 participants; 140 were recruited to allow for attrition.” Everything an examiner needs is in one sentence, including the fact that you thought about drop-out.
What about the rules of thumb?
They have a place — as a sanity check on a power analysis, or where a formal calculation is not feasible — provided you name them rather than presenting them as natural law. The commonly cited ones: for multiple regression, minimum sample sizes that scale with the number of predictors, which is why adding a fifth and sixth predictor to a model is not free; for chi-square, the requirement that expected counts in your contingency table are large enough for the test to behave, which is the real constraint when you cross-tabulate two variables with several categories each and discover cells containing two people; and for factor analysis, substantially larger samples than students expect. Cite the rule you used. “A minimum of 50 + 8m was applied, following the convention for testing multiple correlation” is defensible; “a sample of 80 was considered adequate” is not.
How many participants does a qualitative study need?
Fewer than a quantitative study, and for entirely different reasons — qualitative sampling seeks depth and range of experience, not representativeness of a population. Widely cited ranges in the methodological literature run from roughly six to twelve interviews for a fairly homogeneous group exploring a focused question, upward for heterogeneous samples or multiple comparison groups, with case studies and phenomenological work sometimes justifying single figures and grounded theory typically requiring more.
Treat every one of those numbers as a starting expectation for your ethics application, not as a finding. What you defend at examination is the reasoning: who you sampled, why those people could answer this question, how their diversity was ensured, and what told you the data was sufficient.
Is “saturation” still an acceptable justification?
It is the most common justification in South African qualitative dissertations, and it is worth understanding that it is contested rather than settled. The classic version — you stop when new interviews stop producing new codes or themes — comes from grounded theory and travels imperfectly to other approaches. Braun and Clarke, whose thematic analysis method is used in a very large share of South African qualitative work, have argued explicitly that saturation sits awkwardly with reflexive thematic analysis, because meaning is generated through interpretation rather than discovered and exhausted. An alternative framing that has gained ground is “information power”: the richer and more specific your data and the narrower your aim, the fewer participants you need.
The practical implication is not that you must abandon saturation. It is that you must use it consistently with your stated method, and demonstrate it rather than assert it. “Saturation was reached” is an assertion. “No new codes were generated in the final three interviews; coding of interviews 10 to 13 produced only instances of existing categories” is a demonstration, and it costs one extra sentence.
What will your ethics committee want to see?
A number with a reason, and a plan proportionate to it. Ethics committees query sample sizes in both directions: a sample too small to answer the question wastes participants’ time and consent for no scientific return, and a sample far larger than needed collects more personal information than the study requires — which is a data-minimisation problem under South African privacy law as well as an ethical one. Our guide to ethics clearance at a South African university covers what else the committee checks and how long the process runs; on sample size specifically, state your target, your justification, your recruitment route and your attrition allowance, and the question usually does not come back.
What if you cannot reach your target?
Say so, in writing, in the dissertation. Under-recruitment is extremely common and is not by itself a fatal flaw; concealing it is. The correct handling has three parts: report the sample you actually achieved, report the consequence honestly — a post-hoc note on the power your realised sample gives you, or an acknowledgement that your qualitative range is narrower than planned — and discuss it in your limitations rather than burying it. Examiners are experienced people who know that fieldwork in South Africa runs into gatekeeper delays, participants without airtime, institutions that stop responding in December and a research site that reorganises halfway through your data collection.
What they will not accept is a results chapter written as though the shortfall did not happen. If the shortfall is severe enough to change what you can claim, change the claim — narrowing your conclusion to what your data supports is a sign of judgement, not weakness. Talk to your supervisor the moment recruitment stalls rather than at submission; if your supervisor is not responding, our escalation guide sets out the order to raise it in.
Does an honours research report need the same sample as a master’s dissertation?
No. Scope scales with the level, and so does what is expected of your sample. An honours research report is normally a smaller, more contained study completed within a year, and a modest sample with a clear justification is entirely appropriate. A master’s dissertation is expected to sustain a defensible design; a doctoral thesis is expected to make an original contribution, which usually implies a sample capable of supporting a claim nobody has made before. Our guide to the differences between the three levels sets out the expectations that flow from each.
Can you avoid the recruitment problem entirely?
Sometimes, and it is worth considering seriously before you commit to fieldwork. Analysing existing survey data sidesteps recruitment, ethics complexity around new participants and the cost of getting to a research site — and South Africa has unusually good free microdata available to students, documented in our guide to accessing South African data for your dissertation. Sample size in that world becomes a question of which subgroup you can analyse rather than how many people you can find, and the weights need respecting. It is not the right choice for every question, but for a working student with limited evenings it is the option most often overlooked.
Whichever route you take, the justification has to be written down, in the methodology chapter, in language an external examiner can check. Getting that paragraph drafted while the reasoning is fresh — rather than reconstructing it eighteen months later — is exactly the kind of task Tesify is built to hold: your chapter structure, your decisions and your sources in one project, so the sentence you need at submission is already there.
FAQ: sample size for a dissertation
Is 30 participants enough for a master’s dissertation?
It depends entirely on your design. Thirty may be ample for a qualitative study and badly underpowered for a regression with several predictors. There is no universal minimum in South Africa; there is only a justification your examiner accepts or does not.
What is a power analysis?
A calculation that returns the sample size needed to detect an effect of a given size, at a given significance level, with a given probability. Run before data collection it tells you what to recruit; run afterwards it tells you what your study could realistically have found.
What software do I use for sample size calculation?
G*Power is free, widely taught and the tool most supervisors expect to see cited. Several statistical packages include power modules, and online calculators exist for simple survey designs — cite whichever you used, with the parameters you entered.
What effect size should I assume?
Preferably one taken from a comparable published study in your field. Failing that, a conventional benchmark with a stated argument for why it is appropriate. Never choose the effect size that gives you the sample you wanted.
How many interviews are enough for a qualitative dissertation?
Commonly cited starting ranges run from around six to twelve for a homogeneous group and a focused question, more for diverse samples or multiple groups. The number you defend is the one your sampling logic and your data sufficiency argument support.
How many focus groups do I need?
Enough to cover the range of perspectives your question requires, with each group large enough to generate discussion and small enough that everyone speaks. Justify the number by the segments you needed to hear from, not by a target.
Do I need to calculate sample size for a case study?
No power calculation applies, but you must still justify case selection — why this case or these cases, what makes them informative, and what the boundaries of the case are. Selection logic replaces sample size.
What happens if my sample is smaller than planned?
Report the achieved sample, address the consequence for what you can claim, and discuss it in your limitations. Under-recruitment handled openly is a limitation; under-recruitment concealed is an integrity problem.
Does my sample need to be representative?
Only if you intend to generalise to a population statistically, which requires probability sampling. Much postgraduate research uses non-probability samples and generalises analytically instead — state which you are doing and match your conclusions to it.
Should I over-recruit to allow for drop-out?
Yes for longitudinal or multi-stage designs, and state the allowance in your methodology. Do not over-collect personal information beyond what your study needs, though — collecting more data than the design requires raises both ethical and privacy questions.