Sampling Methods in Research Explained
Probability and non-probability sampling, when each is appropriate, and how to justify your sample size, the choices that decide who your findings apply to.
Your sample determines who your findings apply to. Choosing a sampling method is therefore not a technicality but a decision about generalisability, and markers expect it named, justified and sized with reasoning rather than convenience.
Probability Sampling
Every member of the population has a known, non-zero chance of selection, which supports statistical generalisation.
- Simple random: every unit equally likely, the gold standard, but needs a full sampling frame.
- Systematic: every nth unit from a list, simpler, but beware hidden periodicity.
- Stratified: divide the population into strata (e.g. age bands) and sample within each, improves representativeness for known subgroups.
- Cluster: sample whole groups (e.g. schools), practical for dispersed populations, at some cost to precision.
Non-Probability Sampling
Selection is not random. Generalisation is limited but the approach is legitimate, especially in qualitative research, when justified.
- Purposive: deliberately selecting information-rich cases relevant to the question. The qualitative default.
- Snowball: participants recruit others, for hard-to-reach or hidden populations.
- Convenience: whoever is available, weakest for generalisation. Use only with explicit acknowledgement.
- Quota: non-random filling of preset category targets.
Justifying Sample Size
For quantitative work, size should come from a power calculation based on expected effect size, significance level and desired power, not a round number. For qualitative work, the guiding concept is saturation: sampling until new participants stop yielding new themes (often 12-20 for homogeneous groups). State which logic you used.
Sampling and Bias
Every method carries bias risk: non-response bias, self-selection in convenience samples, coverage gaps in the sampling frame. Naming these threats, and any steps to mitigate them, signals methodological awareness and belongs in your limitations if unresolved.
Checklist
- Have I named the sampling method and matched it to my design?
- Is the sample size justified (power calculation or saturation)?
- Have I acknowledged the relevant bias risks?
- Is the population my findings apply to stated clearly?
- Have I described the sampling frame and its gaps?
- Is my recruitment route described in enough detail to be repeated?
Choosing Between the Probability Methods
The four probability methods are not interchangeable. Each buys something and costs something, and the justification a marker wants is the trade you made.
| Method | Buys you | Costs you | Needs |
|---|---|---|---|
| Simple random | Cleanest basis for inference | Can miss small subgroups by chance | A complete list of the population |
| Systematic | Easy to administer in the field | Bias if the list has a hidden cycle | An ordered list with no periodicity |
| Stratified | Guaranteed representation of known subgroups, better precision | More complex, and you must weight correctly | Knowing the strata in advance |
| Cluster | Feasible across dispersed populations | Larger sample needed for the same precision | Natural groupings, and awareness of clustering in analysis |
Cluster sampling carries a consequence students often miss. If you sample whole schools, pupils within a school resemble each other more than they resemble pupils elsewhere, so the observations are not independent. Standard tests assume independence, so an analysis that ignores the clustering will produce standard errors that are too small and p-values that are too flattering.
Non-Probability Methods, Used Well
Non-probability sampling is not a lesser option. In qualitative research it is the correct option, because the aim is information richness rather than statistical representativeness. What makes it weak is using it by default and then writing as though the sample were representative anyway.
| Method | Strongest use | Main risk |
|---|---|---|
| Purposive | Selecting cases that can actually answer the question | Selection driven by expected answers rather than relevance |
| Snowball | Hidden or stigmatised populations | Samples cluster within one social network |
| Convenience | Pilot studies and instrument testing | Overclaiming generalisability |
| Quota | Fieldwork with no sampling frame available | Non-random selection within each quota cell |
| Theoretical | Grounded theory, sampling driven by emerging concepts | Only coherent if analysis genuinely runs alongside collection |
The Sampling Frame Is Part of Your Method
The sampling frame is the list you actually sampled from, and it is rarely identical to the population you care about. A survey of "university students" drawn from one faculty's mailing list has a frame that excludes most of the population it names. Coverage error of this kind is invisible unless you describe the frame, which is why examiners look for it.
State three things: what the target population is, what list or route you drew from, and who that route systematically leaves out. Doing so costs a short paragraph and pre-empts the obvious criticism.
Justifying Size Without Reaching for a Round Number
| Study type | Basis for the number | What to report |
|---|---|---|
| Experimental or survey | Power calculation | Expected effect size, alpha, power, resulting n, and where the effect size came from |
| Regression | Events or cases per predictor | Number of predictors and the rule of thumb applied |
| Grounded theory | Theoretical saturation | How you judged that new data stopped changing the categories |
| Reflexive thematic analysis | Information power and depth of data | Why the sample suits the scope of the question |
| Case study | Analytic rather than statistical logic | Why these cases, and what they are cases of |
The commonest weak sentence in a methods chapter is a bare assertion that a round number was chosen because it was appropriate. Replace it with the calculation or the principle. Even where practical limits drove the number, saying so honestly and discussing the consequence reads better than a fabricated rationale.
Bias, Named Properly
| Bias | What happens | What helps |
|---|---|---|
| Coverage | The frame omits part of the population | Describe the omission, use multiple frames where possible |
| Non-response | Those who respond differ from those who do not | Report the response rate, compare respondents with the known population |
| Self-selection | Volunteers hold stronger views | Acknowledge it, avoid framing recruitment around the outcome |
| Survivorship | Only those still present are sampled | Consider who left, and why that matters to your question |
| Attrition | Dropouts differ from completers | Report dropout numbers and compare the two groups |
Common Mistakes and Fixes
| Mistake | Fix |
|---|---|
| Calling a convenience sample "random" | Random has a technical meaning. Use it only for probability sampling |
| No sample size justification at all | Give a power calculation, a saturation account, or an honest practical constraint |
| Generalising from a single institution | Bound the claim to the population sampled |
| Ignoring clustering in the analysis | Use multilevel or cluster-robust methods, or state the limitation |
| Reporting recruitment too vaguely to repeat | Say where, when, how, and what the inclusion criteria were |
| Omitting the response rate | Report it. Its absence is read as a poor one |
Discipline Notes
- Health research: power calculations are usually expected, and ethics committees ask for them explicitly.
- Psychology: student samples are common and acceptable, provided the limitation on generalisability is stated rather than glossed.
- Education: cluster structures are almost unavoidable, so plan for them in the analysis rather than discovering them afterwards.
- Business and management: organisational access often dictates the sample. Say so plainly and discuss what it means for the findings.
- Sociology and anthropology: purposive and theoretical sampling dominate, and representativeness is usually the wrong criterion to apply.
Where Ethical Support Fits
Asking a statistics adviser to check a power calculation, or a supervisor whether your recruitment route will pass ethics review, is ordinary research training. Both are the kinds of question these services exist to answer.
What stays yours is recruiting the participants, applying your criteria consistently and reporting the sample as it actually turned out. A smaller sample than planned is a limitation to state, not a number to round upward.
Frequently Asked Questions
Is 30 participants enough?
There is no such number. Thirty appears in textbooks as a rough point at which sampling distributions behave, which is a statistical observation, not a sample size rule. Use a power calculation.
Can I mix sampling methods?
Yes, and it is common. You might stratify by faculty and then sample randomly within each. Describe the full procedure step by step so it can be followed.
What if I cannot get the sample I planned?
Report what you achieved, explain the shortfall, and discuss what it means for your conclusions. Examiners are experienced in recruitment difficulty and respond well to candour.
Does my sampling method need ethics approval?
The study does, and how you recruit is part of what the committee reviews, especially for snowball sampling, vulnerable groups or any route where consent could feel pressured.
How do I justify a sample size in qualitative work?
By reference to the analytic approach. Grounded theory uses saturation. Reflexive thematic analysis uses information power and the depth of the accounts. Say which logic applies to you.
Should the sampling section sit in methods or in limitations?
Both. The method and its justification belong in methods. The consequences you could not resolve belong in limitations, and any that shape the interpretation belong in the discussion.
Your Next Step Today
Write three sentences: who your target population is, what list or route you actually sampled from, and who that route excludes. If the second and third are hard to write, you have found the gap a marker will find first.
Trusted Sources
- Andrade C, Sample Size and its Importance in Research, Indian Journal of Psychological Medicine. Open access. Accessed 11 August 2026.
- University of Southern California, Quantitative Methods. Accessed 11 August 2026.
- University of Southern California, Qualitative Methods. Accessed 11 August 2026.
- University of Southern California, Limitations of the Study. Accessed 11 August 2026.
- UCLA Office of Advanced Research Computing, Choosing the Correct Statistical Test, useful for matching design to analysis. Accessed 11 August 2026.
Sampling expectations differ by discipline and by ethics committee. Where this guide and your department's guidance differ, follow your department.
Related services
Related guides
- Writing the Methodology Chapter: Design, Sampling and Justification
- Reliability and Validity in Research
- Choosing the Right Statistical Test
- Survey Design Basics for Student Research Projects
All academic writing guides or browse our academic writing services.