Probability vs Non-Probability Sampling: Which One Fits Your Study?
Use probability sampling when your study requires population inference and you can define a probability-based selection process. Non-probability sampling can be appropriate for qualitative inquiry, formative research, specialized populations or pilot work where statistical population representation is not the aim. The key question is not which approach sounds more rigorous, but whether the sampling strategy supports the conclusion you intend to make.
Probability sampling: good.
Non-probability sampling: bad.
That is the version many students learn.
It is also far too crude to guide real research.
A purposive sample in a qualitative study may be exactly right.
A convenience sample used to estimate district prevalence may be exactly wrong.
The sampling method has to fit the inference.
What is probability sampling?
Probability sampling uses a probability mechanism to select units from a defined population or sampling frame.
Common probability approaches include:
- simple random sampling;
- systematic sampling;
- stratified sampling;
- cluster sampling;
- multistage probability sampling.
The precise selection probabilities can differ across designs.
What they share is that inclusion is governed by a defined probability process rather than solely by researcher judgement or participant availability.
Probability sampling is particularly useful when you want to make statistical inferences about a target population.
What is non-probability sampling?
In non-probability sampling, selection is not based on a probability mechanism that gives population units known selection probabilities in the way required for probability-based inference.
Examples include:
- convenience sampling;
- purposive sampling;
- quota sampling;
- snowball or chain-referral sampling;
- theoretical sampling in appropriate qualitative methodologies.
These approaches are not one interchangeable category.
Purposive sampling is not simply convenience sampling with a nicer name.
The sampling logic should be tied to the study purpose.
The right sampling strategy is the one that supports the inference you intend to make.
Do not rank sampling methods without first asking what the study needs to conclude.
three studies, three different sampling decisions
Study A: district prevalence of modern contraceptive use
Research question:
What proportion of women aged 15-49 in District X currently use a modern contraceptive method?
You want a population estimate.
Your sampling strategy therefore needs to support population inference.
A probability-based design is likely to be important.
Depending on the setting, that may involve:
- a household sampling frame;
- enumeration areas;
- multistage sampling;
- weighting;
- design-aware analysis.
Recruiting women who happen to attend four clinics would answer a different question.
You would be estimating contraceptive use among the recruited clinic attendees, not necessarily among women across the district.
Study B: experiences of women denied a particular service
Research question:
How do women describe their experiences after being denied a requested contraceptive method?
Now the goal is not to estimate district prevalence.
You want participants with direct experience of the phenomenon.
Purposive sampling may be entirely appropriate.
You may intentionally seek:
- different ages;
- different facilities;
- different reasons for denial;
- variation in subsequent decisions.
A simple random sample of all women in the district could be extremely inefficient because most people may never have experienced the event.
Study C: pilot-testing a new questionnaire
Aim:
To identify comprehension problems and test the feasibility of administering a new 20-item questionnaire.
A convenience sample may be defensible for an early pilot if the goal is practical instrument testing rather than population estimation.
But the conclusion must match:
“The questionnaire was feasible in this pilot group.”
Not:
“The questionnaire is acceptable to all university students in Uganda.”
Same recruitment method.
Very different claim.
Probability sampling is not automatically representative
Researchers often write:
“Probability sampling was used to ensure a representative sample.”
Be careful.
Probability sampling can support principled population inference.
It does not guarantee that every realised sample perfectly represents every characteristic of the target population.
Problems can still arise from:
- incomplete sampling frames;
- differential non-response;
- measurement error;
- poor implementation;
- undercoverage;
- incorrect weights;
- analysis that ignores the design.
Probability sampling is a design feature.
“Representative” is a much larger claim.
Non-probability sampling is not automatically weak
The key is whether the sampling method fits the methodological purpose.
Qualitative research often deliberately seeks information-rich participants.
For example, if you are studying how district health managers make budget-allocation decisions, sampling people because they actually hold those decision-making roles makes methodological sense.
Randomly selecting residents from the district would not somehow make the study more rigorous.
It would make the sampling irrelevant to the question.
You wrote:
“Purposive sampling was used because it is cheaper and easier.”
That may describe why you liked it.
It does not justify it methodologically.
A stronger rationale explains why the selected participants are positioned to illuminate the phenomenon under study.
For example:
Participants were purposively selected because the study required respondents with direct responsibility for district-level planning and budget allocation.
That tells the reader what the sampling is designed to achieve.
Is convenience sampling ever acceptable?
Yes, but its limitations should be explicit.
Convenience sampling may be useful for:
- early piloting;
- instrument pretesting;
- exploratory work;
- feasibility testing;
- classroom or training exercises;
- research where a defined accessible population itself is the target.
The problem starts when a convenience sample is used to support conclusions that require broader population inference.
A survey of 5,000 Instagram followers is not transformed into a national probability sample by sample size.
A large biased sample can be very precisely wrong about the population you wanted to study.
Sample size and sampling validity are separate questions.
What about snowball sampling?
Snowball or chain-referral approaches can help researchers reach populations for whom a conventional sampling frame is unavailable or difficult to construct.
But “snowball sampling” should not become the automatic label for:
“Participant A gave me the phone number of participant B.”
The recruitment process, network structure and inferential limitations still need to be described.
More specialized network-based designs exist and should not be confused with informal referral recruitment.
Can I combine sampling approaches?
Yes.
Real studies often do.
For example:
- randomly select districts;
- randomly select facilities within districts;
- purposively select key informants within selected facilities.
The sampling strategy can differ across study components because the inferential purpose differs.
Similarly, a mixed-methods study may use:
- probability sampling for a population survey;
- purposive sampling for follow-up qualitative interviews.
That is coherent if each component answers a different part of the research question.
How do I choose between probability and non-probability sampling?
Ask these questions.
1. What conclusion do I want to make?
Population prevalence?
Group comparison?
Experience?
Mechanism?
Instrument feasibility?
2. What is my target population?
Can it be clearly defined?
3. Do I have a usable sampling frame?
If not, can one be constructed?
4. Does every eligible unit need a probability of selection for my inference?
If you need design-based population estimates, that becomes important.
5. Do I instead need information-rich cases?
If the aim is qualitative depth, purposive logic may fit better.
6. What limitations will the sampling place on my conclusion?
Write these before data collection.
Not after the reviewer asks.
What do I actually write in my methods section?
Probability example
A two-stage probability-sampling design was used. Enumeration areas were selected with probability proportional to size, followed by random selection of households within selected areas. The analysis accounted for the sampling design and applicable weights.
Only use that wording if it describes your actual procedure.
Purposive example
Participants were purposively selected to capture variation in facility type, professional role and direct experience with contraceptive-service delivery.
Convenience example
A convenience sample of eligible students attending the pilot session was recruited to assess questionnaire comprehension and administration procedures. Findings from the pilot were not intended to estimate population-level acceptability.
Notice the final sentence.
It aligns the claim with the sample.
Finish this sentence:
I need this sample to allow me to conclude that...
Then inspect your sampling strategy.
If you need to conclude something about a population, ask whether your selection method supports that inference.
If you need depth about a specific experience, ask whether your sample deliberately includes people who can illuminate it.
If your sampling method cannot support the conclusion, change one of them now.
Frequently asked questions
- Is probability sampling always better?
No. It is particularly useful for population inference, but purposive or other non-probability approaches may be more appropriate for different research aims.
- Is purposive sampling biased?
Purposive sampling is intentionally selective. In qualitative research, the aim is often information richness rather than statistical representativeness. The important issue is whether the selection logic fits the study.
- Can convenience sampling be used in quantitative research?
Yes, but the inferential limitations must be clear. A convenience sample generally does not provide the same basis for population inference as a probability sample.
- Does a large non-probability sample become representative?
Not automatically. Increasing sample size does not fix systematic selection or coverage problems.
- Can mixed-methods studies use two different sampling strategies?
Yes. A probability-sampled quantitative component and purposively sampled qualitative component can be entirely coherent when they serve different purposes.
Use the Methods Bench Study Design Decision Guide to clarify the inference your study needs before choosing a sampling approach.
If your study requires a probability-based sample-size calculation, continue to the Methods Bench Sample Size Calculator.
References and further reading
- 1.World Health Organization. WHO STEPS surveillance guidance on preparing probability samples.
- 2.CDC/NCHS. National Health Interview Survey methods and probability-sampling documentation.
- 3.JBI methodological guidance on sampling, applicability and evidence generation.
- 4.Methods Bench. Simple Random Sampling Explained With a Real Research Example.
Take it further
Better research, once a week.
Practical methods guidance, research tools and funding opportunities from Methods Bench.
Get practical research notes and new opportunities
Short, useful emails for researchers in Uganda and East Africa. No spam.