You almost never study an entire population. Instead you study a sample — a subset — and reason back to the whole. How you choose that sample is one of the most scrutinized parts of your methodology, because it determines how far your findings can be trusted to apply beyond the people you actually studied. Choose a sample badly and even flawless analysis produces misleading conclusions.
This guide explains the main sampling techniques, splitting them into the two great families — probability and non-probability — with definitions, examples, and guidance on when each is appropriate.
Key terms first
- Population: the entire group you want to draw conclusions about (e.g. all nurses in Ghana).
- Sampling frame: the actual list from which you draw (e.g. a registry of licensed nurses).
- Sample: the subset you actually study.
- Generalizability: how confidently your sample's results apply to the population.
The central question every sampling method answers is: how do you pick who gets in? — and whether that picking is random or deliberate.
Probability sampling
In probability sampling, every member of the population has a known, non-zero chance of selection. Because selection is random, you can estimate sampling error and generalize to the population with statistical confidence. Probability methods are the backbone of quantitative research.
Simple random sampling
Every member has an equal chance of selection — like drawing names from a hat or using a random-number generator on your sampling frame. It is the purest form and the benchmark others are compared to. Requires a complete sampling frame, which is not always available.
Systematic sampling
You select every k-th member from an ordered list (e.g. every 10th patient on a register), after a random start. Easier to administer than simple random sampling, and usually just as representative — provided the list has no hidden periodic pattern that lines up with your interval.
Stratified sampling
You divide the population into meaningful subgroups (strata) — say, by region or age — and sample randomly within each, often in proportion to their size. This guarantees representation of key subgroups and improves precision when the strata differ.
Cluster sampling
You divide the population into naturally occurring groups (clusters) — schools, villages, clinics — randomly select whole clusters, and study everyone (or a random sample) within the chosen clusters. Far cheaper and more practical for geographically dispersed populations, at some cost to precision. Often the only feasible option for large national studies.
Non-probability sampling
In non-probability sampling, selection is not random — some members have no chance of being chosen, and you cannot calculate sampling error. These methods cannot statistically generalize to a population, but they are entirely appropriate for qualitative research, pilot studies, and hard-to-reach populations.
Convenience sampling
You select whoever is easiest to reach — students in your class, patients in the waiting room, passers-by. Quick and cheap, but prone to bias, so generalize from it only with caution. Common in early-stage and exploratory work.
Purposive (judgmental) sampling
You deliberately select participants who fit specific criteria relevant to your question — for example, only nurses with more than ten years' experience. This is the workhorse of qualitative research, where you want information-rich cases rather than a statistically representative cross-section.
Snowball sampling
Existing participants refer others, growing the sample like a rolling snowball. Invaluable for hidden or stigmatized populations (e.g. people in informal trades) where no sampling frame exists. The trade-off is that referrals tend to share characteristics, narrowing the sample.
Quota sampling
You set targets ("quotas") for subgroups — say, 50 men and 50 women — and fill them by convenience. It mimics stratified sampling's coverage without the random selection, so it is faster but not statistically representative.
Probability vs non-probability: which to use?
| | Probability | Non-probability | |---|---|---| | Selection | Random | Deliberate / convenient | | Generalizable? | Yes, statistically | No (transferable, not generalizable) | | Best for | Quantitative, descriptive surveys | Qualitative, exploratory, hard-to-reach groups | | Needs a sampling frame? | Usually yes | No | | Example methods | Simple random, systematic, stratified, cluster | Convenience, purposive, snowball, quota |
The right choice depends on your approach and your constraints. If your study is quantitative and you need to generalize, aim for a probability method and a defensible sample size. If your study is qualitative and you want depth, purposive sampling is usually the principled choice — and you justify it on the grounds of richness, not representativeness.
How big should the sample be?
There is no universal number. For quantitative studies, sample size is driven by the precision you need, the expected effect size, and your statistical power target — often calculated with a power analysis. For qualitative studies, the guiding idea is saturation: you keep sampling until new participants stop adding new themes. Whatever you choose, state and justify it — an unjustified sample size is a red flag to examiners.
Tie sampling back to your methodology
Sampling does not stand alone. It must align with your research design and approach: an experiment implies probability sampling and statistical analysis; an in-depth case study implies purposive sampling and qualitative analysis. Keeping that whole chain consistent is what makes a methodology convincing.
PaceReseacher's Methodology Copilot helps you choose a sampling strategy that fits your design, articulate and justify your sample size, and write the sampling section with real methodological citations. Return to the methodology hub to see how sampling slots into the full chapter.