Sampling methods are the techniques researchers use to select a subset of individuals (a sample) from a larger group (the population) so that conclusions drawn from the sample can be applied to the whole. There are two broad families of sampling method in research: probability sampling, where every member of the population has a known, non-zero chance of selection (enabling statistical generalisation), and non-probability sampling, where selection is based on convenience or judgement (useful for exploratory or qualitative work).
Choosing the right sampling method is one of the most consequential decisions in your dissertation. It determines whether your findings genuinely represent the population, how much sampling error you carry, and which statistical tests you are entitled to run. This guide explains each method, when to use it, and how to calculate a defensible sample size.
Population, sample and sampling frame: the three terms to fix first
Before you pick a sampling method, you must be precise about three related but distinct ideas. Confusing them is the single most common error in a methodology chapter, and examiners notice it immediately.
- Population (target population): the entire set of units you want your conclusions to apply to — for example, “all full-time undergraduate students at UK universities in 2026”. The population is defined by your research question, not by what is convenient.
- Sampling frame: the actual, accessible list from which you draw your sample — for example, a university’s student-record database or an electoral register. The frame is your operational stand-in for the population, and any gap between the two (people missing from the list, duplicates, out-of-date entries) introduces coverage error.
- Sample: the specific units you actually study. A good sample is large enough and selected in a way that lets you generalise back to the population with a quantified level of confidence.
The logical chain runs population → sampling frame → sample. Probability sampling can only deliver its promise of representativeness when the sampling frame closely matches the population. For a fuller treatment of how these levels relate, see our guide on population vs sample.
The two families: probability vs non-probability sampling
Every sampling method belongs to one of two families, and the family you choose flows directly from your research aim.
- Probability (random) sampling gives every unit a known, non-zero chance of selection. Because selection is random, you can estimate sampling error, build confidence intervals, and generalise to the population. This is the backbone of quantitative, survey-based and experimental research.
- Non-probability sampling selects units by convenience, quota or researcher judgement. You cannot calculate the probability of selection, so statistical generalisation is not defensible — but these methods are efficient, cheap and entirely appropriate for qualitative, exploratory and hard-to-reach populations.
“In probability sampling, each case in the population has a known and usually equal chance of being selected. This is most commonly associated with survey and experimental research strategies.” (Source: Saunders, Lewis & Thornhill, Research Methods for Business Students)
Probability sampling methods (with examples)
1. Simple random sampling
How it works: Every unit in the sampling frame is assigned a number, and units are drawn purely by chance — using a random number generator, a lottery, or a spreadsheet’s random function. Each unit has an identical probability of selection and each combination is equally likely.
2. Systematic sampling
How it works: You select every k-th unit from an ordered list after a random start. The sampling interval k = N ÷ n, where N is the population size and n is the desired sample. It is faster than simple random sampling and easy to administer, but it fails if the list has a hidden periodic pattern that coincides with k.
3. Stratified sampling
How it works: The population is divided into mutually exclusive strata (subgroups that share a relevant characteristic, such as year of study, gender or department), and you sample randomly within each stratum. In proportionate stratified sampling, each stratum contributes in proportion to its share of the population, which improves precision when the strata differ on your outcome.
4. Cluster sampling
How it works: The population is split into naturally occurring groups (clusters — schools, hospitals, postcodes), and you randomly select whole clusters, then study every unit within the chosen clusters. It is highly cost-effective for geographically dispersed populations, though clusters that are internally homogeneous inflate the sampling error.
5. Multistage sampling
How it works: A practical hybrid that combines several probability stages. You sample clusters, then sub-sample within them, repeating across stages. National surveys rely on it: random regions, then random towns, then random households, then one random adult per household.
Non-probability sampling methods (with examples)
1. Convenience sampling
How it works: You recruit whoever is easiest to reach — passers-by, your own course mates, online volunteers. It is quick and cheap but carries the highest risk of bias, because the easy-to-reach rarely represent the whole population.
2. Purposive (judgemental) sampling
How it works: You deliberately select information-rich cases that fit your criteria, using your knowledge of the population. It is the workhorse of qualitative research, where the goal is depth and relevance, not statistical representativeness.
3. Quota sampling
How it works: The non-random cousin of stratified sampling. You set quotas for key subgroups (e.g. 50 men, 50 women) and fill each quota by convenience until it is met. It controls the composition of the sample without random selection.
4. Snowball sampling
How it works: Existing participants refer further participants, growing the sample like a rolling snowball. It is invaluable for hidden, stigmatised or hard-to-reach populations where no sampling frame exists.
Master comparison table
Use this table to match a method to your design at a glance.
| Method | How it works | Pros | Cons | Best for |
|---|---|---|---|---|
| Simple random | Random draw from a numbered frame | Unbiased; easy to analyse | Needs a full list; can miss small subgroups | Small, well-listed populations |
| Systematic | Every k-th unit after a random start | Fast; even spread | Fails on periodic lists | Long ordered lists |
| Stratified | Random within proportional subgroups | High precision; guarantees subgroup cover | Needs known strata; more admin | Heterogeneous populations |
| Cluster | Random whole groups, study all within | Cheap for dispersed populations | Higher sampling error | Geographically spread populations |
| Multistage | Several probability stages combined | Practical at national scale | Complex; compounding error | Large national surveys |
| Convenience | Whoever is easiest to reach | Fast; free | High bias; not generalisable | Pilots, quick exploratory work |
| Purposive | Deliberate information-rich cases | Deep, relevant data | Researcher bias; subjective | Qualitative depth studies |
| Quota | Convenience within fixed subgroup quotas | Controlled composition; quick | Non-random; selection bias | Market research, fieldwork |
| Snowball | Participants refer further participants | Reaches hidden populations | Network bias; not representative | Hard-to-reach, stigmatised groups |
How to choose a sampling method: a step-by-step process
Work through these questions in order; each answer narrows your options.
- What is your aim? If you need to generalise to a population with quantified confidence, you need probability sampling. If you want depth, meaning or to explore a phenomenon, non-probability sampling is legitimate and often preferable.
- Do you have a sampling frame? Probability methods require a usable list. No frame (e.g. homeless adults, gig workers) usually forces a non-probability route such as snowball or purposive sampling.
- How is the population distributed? Heterogeneous on a key variable → stratify. Geographically scattered with cost limits → cluster or multistage.
- What are your resources? Time, budget and access constrain everything. Cluster and quota sampling exist largely to control cost.
- What analysis will you run? Inferential statistics and significance testing assume probability sampling; thematic or content analysis does not. Align the method with your planned data-collection and analysis approach from the outset.
Worked calculation: how big should your sample be?
Sample size is a calculation, not a guess. For a survey estimating a proportion in a large population, the standard formula is n = z² · p(1 − p) ÷ e², where z is the z-score for your confidence level, p is the expected proportion, and e is the margin of error you will tolerate.
Suppose you want 95% confidence, the most cautious proportion (p = 0.5, which maximises the sample), and a ±5% margin of error (e = 0.05).
Step 1 — set the values: z = 1.96 (for 95% confidence), p = 0.5, e = 0.05.
Step 2 — numerator: z² · p(1 − p) = (1.96)² × 0.5 × (1 − 0.5) = 3.8416 × 0.25 = 0.9604.
Step 3 — denominator: e² = (0.05)² = 0.0025.
Step 4 — divide: n = 0.9604 ÷ 0.0025 = 384.16.
Step 5 — round up: you cannot survey a fraction of a person, so n ≈ 385 respondents.
Now apply systematic sampling to reach them. If your sampling frame contains N = 20,000 students and you need n = 385, the sampling interval is:
k = N ÷ n = 20,000 ÷ 385 = 51.9, which you round down to k = 51.
Pick a random start between 1 and 51 — say 8 — then select students 8, 59, 110, 161, … taking every 51st name until you have 385.
Two practical notes. First, if your population is small, apply the finite population correction, which reduces the required n. Second, always recruit more than 385 to absorb non-response: if you expect a 50% response rate, send the survey to roughly 770 people. For help running the numbers and the subsequent tests, our statistical analysis service can support your design.
Sampling error and sampling bias
Two threats undermine a sample, and they are not the same thing.
- Sampling error is the natural, random difference between a sample statistic and the true population value, simply because you studied a subset rather than everyone. It is unavoidable but measurable — it shrinks predictably as sample size grows, and it is what your margin of error and confidence interval quantify.
- Sampling bias is a systematic error that makes the sample unrepresentative in a consistent direction. Unlike sampling error, a bigger sample does not fix it. Common sources include a flawed sampling frame (coverage bias), self-selection (volunteers differ from non-volunteers) and non-response (those who decline differ from those who reply).
The defining advantage of probability sampling is that it controls bias by design and lets you quantify the remaining error. To check that your instrument measures consistently and accurately alongside a sound sample, review our guidance and run your survey carefully — see how to conduct surveys.
Common mistakes to avoid
- Confusing the population with the sampling frame, then claiming generalisability the frame cannot support.
- Calling a convenience sample “random” — sharing a link widely is not random selection.
- Running inferential statistics on a non-probability sample and reporting p-values as if they generalise.
- Choosing a sample size by rule of thumb (“30 is enough”) instead of calculating it from confidence, proportion and margin of error.
- Ignoring non-response: a perfectly drawn sample becomes biased if only one type of person replies.
- Using systematic sampling on a list with a hidden cycle that matches your interval k.
Need help designing your sampling strategy?
Our experts can shape your methodology, justify your sample and write a distinction-level dissertation.
Bringing it together
Sampling is where the credibility of your whole study is won or lost. Fix your population and sampling frame, decide whether your aim demands probability or non-probability sampling, pick the specific method that fits your population’s structure and your resources, and calculate — never guess — the sample size you need. Get that chain right and your findings will stand up to the toughest examiner.
Related methodology guides
- Pilot Study
- Triangulation in Research
