What is sampling in research?

Choosing a subset of a population to study, because it is usually impossible to study everyone. The way you choose determines what you can claim about the wider population.

What is the difference between probability and non-probability sampling?

In probability sampling, every member of the population has a known chance of being selected, which supports statistical generalization. In non-probability sampling, selection is based on convenience or judgment, so results cannot be generalized in the same way.

How big should my sample be?

It depends on the design. Quantitative studies often use power analysis or guidelines for the planned analysis, while qualitative studies use smaller samples chosen for depth and continue until little new information emerges.

Is convenience sampling acceptable?

It is common in student projects, but it limits generalization. If you use it, be honest about the limitation and describe your sample clearly.

What is a sampling frame?

The list or source from which you select your sample, such as a class roster or a registry. If the frame misses part of the population, your sample will too.

Sampling Methods Explained: Probability and Non-Probability

You almost never study everyone you care about. A survey of every nurse in the country, or every first-year student in the world, is out of reach, so researchers study a sample and draw conclusions about a wider population. How you choose that sample determines what those conclusions are worth.

This guide explains the key terms, the main sampling methods and the biases to watch for, and shows you how to match a method to your project.

Key terms

Term	Meaning
Population	The whole group you want to learn about
Sample	The part of the population that you actually study
Sampling frame	The list or source from which the sample is drawn
Sampling unit	The thing you sample, such as a person, a class or a hospital
Representative sample	A sample that mirrors the population on the features that matter
Sampling bias	A systematic difference between the sample and the population
Response rate	The share of those invited who take part

A common mistake is to describe the sample as if it were the population. Say who you studied, and who you hope to speak about.

Two families of methods

Sampling methods fall into two families.

Probability sampling: every unit has a known, nonzero chance of selection. It supports statistical inference about the population.
Non-probability sampling: selection depends on convenience or judgment. It cannot support statistical generalization, but it may suit exploratory or qualitative work.

Probability sampling

Simple random sampling

Every member of the sampling frame has an equal chance of selection, as in drawing names from a hat or using a random number generator. It is simple and unbiased, but it requires a complete list of the population.

Systematic sampling

You select every nth person from a list, after a random start. For example, from a roster of 1,000 students, you choose every tenth person. It is easy, but it can be biased if the list has a hidden pattern.

Stratified sampling

You divide the population into groups, called strata, such as year of study, and sample from each. This ensures that important subgroups are represented and can improve precision. It requires knowing the characteristics in advance.

Cluster sampling

You divide the population into natural groups, such as schools or hospitals, randomly select some clusters and study everyone, or a sample, within them. It is practical when a full list is unavailable or participants are spread out, though it can be less precise than simple random sampling.

Multistage sampling

A combination: for example, randomly select districts, then schools within districts, then classes within schools, then students within classes. Large national surveys often use it.

Method	How it works	Good for	Watch out for
Simple random	Equal chance for all	Small, well-listed populations	Needs a full list
Systematic	Every nth unit	Large ordered lists	Hidden patterns
Stratified	Sample within groups	Ensuring subgroup representation	Needs group data in advance
Cluster	Sample whole groups	Dispersed populations	Lower precision
Multistage	Several levels	Large surveys	Complexity

Non-probability sampling

Convenience sampling

You choose people who are easy to reach, such as classmates, colleagues or social media contacts. It is quick and cheap and very common in student projects, but it can be heavily biased and cannot represent a wider population.

Purposive or judgment sampling

You deliberately choose participants who have particular experience or characteristics relevant to your question, such as nurses who have mentored new graduates. It is typical in qualitative research, where depth matters more than representation.

Snowball sampling

Participants refer you to others. It is useful for hard-to-reach groups, such as people in sensitive situations, but it tends to reach connected people and may miss isolated ones.

Quota sampling

You set quotas for groups, such as equal numbers of men and women, and fill them non-randomly. It looks like stratified sampling, but selection within groups is not random, so it lacks the statistical basis.

Self-selection

Participants volunteer in response to an open call, for example, an online survey link. It is easy, but those who respond may differ from those who do not.

Choosing a method

Ask these questions.

What do I want to claim? If you want to generalize statistically, you need probability sampling. If you want to understand experiences in depth, purposive sampling may be better.
Is there a sampling frame? Without a list, random methods are hard.
How accessible is the population? Hard-to-reach groups may need snowball methods.
What are my time and resources?
What does my approach and supervisor expect?

Our guide to qualitative vs quantitative research describes how the approach shapes sampling.

Sample size

There is no single answer.

Quantitative studies: the size depends on the analysis, the effect you want to detect and the precision you need. A power analysis, using software or an online calculator, estimates the sample needed to detect an effect of a given size. Rules of thumb exist, but they vary by method. In surveys, a larger sample reduces the margin of error, but with diminishing returns, and representativeness matters as much as size.
Qualitative studies: samples are smaller and chosen for richness of information. Researchers often continue until new interviews add few new themes, a point called saturation. Reported sizes vary widely by design.
Student projects: a realistic target given your time and access. Be honest about what a small sample can show.

A big biased sample is not better than a smaller well-chosen one. Size does not fix bias.

Common biases

Bias	What happens	How to reduce it
Selection bias	The sample differs systematically from the population	Use random methods or document who is excluded
Non-response bias	Those who do not respond differ from those who do	Follow up, keep surveys short, compare respondents with known population features
Volunteer bias	People who opt in are different	Be cautious in generalizing
Survivorship bias	You only see those who remain	Include dropouts where possible
Undercoverage	The frame misses part of the population	Choose a more complete frame

Report your response rate, and discuss how bias may affect your results.

Writing about your sample

In your methodology chapter, describe:

The population and the sampling frame.
The sampling method and why you chose it.
The inclusion and exclusion criteria.
How you recruited participants.
The sample size, and how you decided it.
The response rate or number who completed.
Key characteristics of participants, without breaking anonymity.
Limitations of the sample.

Our guide to the methodology chapter shows where this fits.

A short example

A student wants to understand study habits among first-year students at her university. A random sample would require a list of all first-years from the registrar, which she asks for, though privacy rules mean she can only receive a mailing through the university.

She decides on a stratified approach, inviting equal numbers from each faculty through official email. Response is 28 percent, so she compares respondents with university figures and notes that students in one faculty are underrepresented.

In her limitations section, she says her results describe responding students and may not generalize to non-responders. The method is imperfect, but she described it honestly, which is the most important thing.

(The example is illustrative.)

Ethics and sampling

Recruitment raises ethical issues: people must give informed consent, may decline without penalty and should not feel pressured. Be careful when recruiting your own students, employees or patients, since they may feel obliged to take part. Protect the identity of participants, especially in small or sensitive samples. Our guide to research ethics and ethics approval covers these points.

Common mistakes

Describing a convenience sample as representative.
Choosing the method before the question.
Ignoring the response rate.
Using a tiny sample to make strong claims.
Not describing recruitment.
Confusing quota and stratified sampling.
Forgetting to explain exclusions.
Equating a large sample with a good one.

A quick checklist

Have I defined my population and frame?
Does my method fit what I want to claim?
Have I justified my sample size?
Do I know the likely biases?
Have I planned recruitment and consent?
Will I report response rate and limitations?

Final thoughts

Sampling is where research meets the real world: you cannot study everyone, so you choose carefully and say honestly what your choice allows. Pick a method that fits your question, aim for the best coverage you can manage and describe your sample and its limits clearly. Readers can accept a modest sample that is described honestly, but not one that claims too much.