A sampling strategy is a choice of technique plus a written justification, and examiners assess the justification more closely than the technique. The choice follows from what your question needs — representativeness or information richness — not from what is convenient, and the paragraph has to say which it needed and why the technique delivers it.
Eight steps follow, with five justification paragraphs written out at the length they occupy in a Master’s dissertation. How many participants you need is a separate question with its own conventions, covered in our guide to sample size for postgraduate health research; selecting a small number of cases for comparison is different again, and set out in our piece on case selection for a political science or IR thesis. This one is about which technique, and how to defend it on paper.
Step 1: Decide what your question actually needs
Everything downstream follows from this and it takes one sentence. If your question asks how much, how many, how often, or whether a difference holds in a population, you need representativeness, and representativeness is what probability sampling buys. If your question asks how, why, or what it is like, you need information richness — participants chosen because they have most to say about the phenomenon — and that is what non-probability sampling buys.
Getting this backwards produces the two commonest design faults at Master’s level: a survey of whoever answered, presented as though it described a population, and a set of interviews with people chosen at random, who turn out to have nothing in particular to say. Write the sentence “this study requires [representativeness / information richness] because the research question asks …” before you choose anything.
Step 2: Define the population, then look for a sampling frame
The population is everyone your conclusions are meant to apply to, defined by attribute and by boundary: not “nurses” but “registered nurses working in adult acute inpatient settings in [region], in post at the time of data collection”. The sampling frame is a list of them that you can actually reach.
Master’s candidates routinely discover there is no frame. Professional registers may be partial or inaccessible, organisations may not release staff lists, and the population may not be enumerable at all — people who have experienced something rarely appear on a list. This is not a failure; it is a finding about your design, and it decides the technique. A study without a frame cannot use probability sampling, and saying so plainly is stronger than implying a representativeness you did not have.
| Situation | What it permits | What to write |
|---|---|---|
| Complete accessible frame | Any probability technique | Name the frame, its source and its date |
| Partial frame, known coverage gap | Probability sampling with a stated coverage limitation | Say who is missing from the frame and in which direction that biases results |
| No frame, population enumerable in principle | Non-probability; quota if you know the population structure | State why no frame exists and what you used instead |
| No frame, hidden or hard-to-reach population | Purposive, snowball, theoretical | Justify the access route and its effect on who you reached |
Step 3: Choose the technique
Nine techniques cover almost all postgraduate work. The column that matters is the last one, because it is what your limitations section will have to concede.
| Technique | Family | Use when | Limitation it imports |
|---|---|---|---|
| Simple random | Probability | Complete frame, homogeneous population | May under-represent small subgroups by chance |
| Systematic | Probability | Complete frame, no periodicity in its ordering | Hidden ordering in the list biases the sample |
| Stratified | Probability | Subgroups matter and their proportions are known | Requires the stratifying variable to be on the frame |
| Cluster | Probability | Population is geographically or organisationally grouped | Larger error than simple random for the same number |
| Purposive | Non-probability | Specific characteristics are required for the question | No claim to statistical generalisation |
| Quota | Non-probability | Population structure known, no frame available | Selection within each quota is not random |
| Snowball | Non-probability | Hidden populations, access through existing participants | Samples networks, so isolated members are missed |
| Convenience | Non-probability | Rarely defensible alone; acceptable for a pilot | Unknown and unquantifiable bias |
| Theoretical | Non-probability | Grounded theory, where analysis directs the next recruitment | Cannot be fully specified in advance, which ethics committees query |
Two techniques are commonly misused. Convenience sampling is often what happened, described as purposive; the difference is whether you can name the characteristic you selected on, and if you cannot, it was convenience. And snowball sampling is frequently used without acknowledging that it samples a network — if your participants all know each other, that is a property of your data, not an accident.
Step 4: Write the justification paragraph
The paragraph has four moves: what the question required, which technique delivers it, how the technique was executed, and what it costs. Five worked specimens follow, written at Master’s length. The bracketed prompts mark where your own detail goes.
Stratified random, education. “The research question asks whether attainment differs by school phase, so the study requires a sample representative of both phases in [local authority]. A stratified random sample was drawn from the authority’s published list of maintained schools [date], stratified by phase in proportion to the population, with schools selected within each stratum using a random number generator. Stratification was chosen over simple random sampling because primary schools outnumber secondary schools in the authority by roughly [ratio], and a simple random draw risked too few secondary cases to support a comparison. The frame excludes independent schools, so findings do not extend to that sector.”
Purposive, health. “The question concerns how ward staff adapt [protocol] in practice, which requires participants who have used it rather than a representative cross-section of staff. Participants were selected purposively against three criteria: current employment on a ward operating the protocol, at least [n] months of experience with it, and a role that includes discretion over its application. Recruitment continued until the criteria were filled across [n] wards. The sample supports analytical rather than statistical generalisation, and the findings describe practice under these conditions rather than in the service as a whole.”
Quota, business. “No accessible list of [population] exists, so probability sampling was not available. A quota sample was constructed using published [source] figures for the distribution of firms by size band and sector, with targets set proportionally and recruitment continuing within each cell until filled. Quota sampling reproduces the population’s structure on those two dimensions but not its variation on any other, and selection within cells was not random, so the sample cannot support inference to the population parameters.”
Snowball, social policy. “The population is not enumerable: [characteristic] is not recorded administratively and the group is not organisationally bounded. Recruitment therefore began with [n] participants identified through [gatekeeper organisation], each of whom was asked to refer others meeting the inclusion criteria. Three separate chains were initiated deliberately, from unconnected starting points, to reduce the tendency of a single chain to sample one network. Participants who are isolated from any such network remain unreachable by this method, and the analysis treats the sample as describing connected members of the population.”
Theoretical, grounded theory. “Consistent with grounded theory, sampling was directed by the developing analysis rather than fixed in advance. An initial [n] participants meeting [criterion] were recruited; subsequent recruitment targeted participants whose characteristics were expected to extend or challenge the emerging categories, specifically [example]. The ethics application specified this procedure and an upper limit of [n] participants, which was agreed in advance. Recruitment ended when further cases produced no new properties of the core categories.”
Step 5: Name the limitation before the examiner does
Every technique imports one, and the fourth move of the justification paragraph is where you concede it. A conceded limitation is a sign of methodological judgement; the same limitation discovered by an examiner in a paragraph that claimed more than it could deliver is a correction. The specific pairing to avoid is a non-probability sample described with the vocabulary of representativeness — “the sample was representative of”, “the findings generalise to” — which no purposive or convenience sample supports.
The honest alternative is analytical generalisation: your findings extend to theoretical propositions rather than to populations. If your study has a conceptual framework, this is where its boundary conditions earn their place, because they already state the population and setting within which you claimed anything at all.
Step 6: Plan recruitment, and expect ethics to constrain it
Your sampling strategy is a claim about who you can reach, and the ethics application is where that claim is tested. Gatekeeper permission, the route by which potential participants first hear about the study, whether you may approach people directly, and what you may offer them are all decided before recruitment begins, and each can force a change of technique. Our guide to ethics approval for sociology and social work research covers where university committee approval ends and a separate health-service approval begins.
Build the dependency into the timetable. A technique that requires a gatekeeper you have not approached is not yet a strategy, and the delay between submitting an ethics application and recruiting is the single most common reason a Master’s timetable slips.
Step 7: Write up the sample you achieved, not the one you planned
The methodology chapter describes the strategy; the findings report the sample that resulted, and the two will differ. Give the planned approach, the number approached, the number who participated, and the reasons for non-participation where you know them. If the achieved sample departs from the plan — a stratum under-filled, a quota cell unmet, a chain that stopped — say so in the methodology chapter and carry the consequence into the limitations.
Examiners respond well to this and badly to its absence, because a chapter that describes only the plan invites the question of what actually happened. The full sequence of a methods chapter, and where the sampling section sits inside it, is set out in our guide to writing the methodology chapter.
Step 8: Check it against the question one last time
Read your research question and your justification paragraph together. The question names a population or a phenomenon; the paragraph should name the same one. If the question says “in UK secondary schools” and the paragraph describes one academy trust, either the question is too wide or the sample is too narrow, and the fix belongs in whichever came second. This check is the same discipline that keeps a problem statement aligned with the study attached to it, and it takes five minutes.
How a PhD differs
The reasoning is identical; the standard of defence is higher and the scale is larger. A doctoral thesis is expected to justify the technique against its nearest alternative explicitly, to demonstrate awareness of the sampling literature in its own field rather than in general, and to address non-response or attrition as an analytic issue rather than a footnote. Multi-phase and mixed-methods designs also have to explain how the samples relate: whether the qualitative phase samples from the quantitative one, and on what basis participants were selected from it.
Getting the justification written while the reasoning is fresh
The sampling justification is easiest to write at the moment you make the decision and hardest to reconstruct a year later, when you can remember what you did but not why you ruled out the alternative. Write the four moves as four sentences the day you choose. Draft your dissertation in Tesify and keep the strategy, the recruitment log and the methodology chapter in one workspace, so that the paragraph you submit is the reasoning you actually used. The design stays yours; there is a free plan, and the useful first step is the single sentence about whether your question needs representativeness or information richness.
Frequently asked questions
What is the difference between a sampling strategy and a sample size?
The strategy is how participants are selected; the size is how many. They are justified separately and by different reasoning — the strategy by what the question needs, the size by power calculation, saturation or an established convention in the field.
Is convenience sampling ever acceptable in a Master’s dissertation?
It is defensible for a pilot study or where access is genuinely the binding constraint, provided you call it convenience sampling rather than dressing it as purposive, and provided the limitations section concedes that the bias is unknown and unquantifiable. It is not acceptable as an unexamined default.
How do I know whether I used purposive or convenience sampling?
Ask whether you can name the characteristic you selected on and state the criteria in advance. If you can, it was purposive. If your participants were simply those who were available and willing, it was convenience, whatever the criteria you wrote afterwards.
What is a sampling frame and what if I do not have one?
It is a reachable list of the population. If none exists, say so explicitly and use a non-probability technique: quota if you know the population’s structure from a published source, purposive if specific characteristics matter, snowball if the population is hidden.
Can I use probability sampling for a qualitative study?
You can, but it rarely serves the question. Qualitative work wants participants with most to say about the phenomenon, and a random draw selects without regard to that. The exception is where you want to avoid a suspicion of cherry-picking within an already narrow population.
How many chains should a snowball sample use?
More than one, from unconnected starting points, because a single chain samples a single network. State how many chains you initiated and where each began; that detail is what distinguishes a designed snowball sample from a series of referrals.
Does my sampling strategy have to be fixed before ethics approval?
The procedure does, even where the specific participants are not determined in advance. Theoretical sampling is approvable if the application explains that recruitment will be directed by the analysis and states an upper limit; what committees will not approve is an open-ended plan with no stated bound.
What do I write if my achieved sample is smaller than planned?
State the planned approach, the number approached, the number achieved and the reasons for non-participation, then carry the consequence into the limitations. A shortfall reported and analysed is a methodological observation; a shortfall left unmentioned is a gap an examiner will find.
Can findings from a purposive sample be generalised?
Not statistically, to a population. They can support analytical generalisation to theoretical propositions — that is, your findings extend the framework rather than describing a population. Use that language, and state the boundary conditions within which the claim holds.