Sampling Methods in Research: Types and Examples
Sampling determines who or what contributes evidence to a study. A researcher may define a strong question, choose an appropriate design, and build an excellent instrument, yet still reach misleading conclusions if the sample does not match the population or purpose of the study. The choice of sample therefore affects what the findings can reasonably describe, compare, explain, or generalize.
The central decision is not simply “How many participants do I need?” It is “Which cases need a chance to enter the study, how will they be selected, and what claims will that selection process support?” Some projects need a probability sample because the goal is to estimate characteristics of a wider population. Others deliberately select information-rich cases because the goal is depth, diversity of experience, or theoretical insight rather than population estimation.
This guide explains the main sampling methods in research, how probability and non-probability approaches differ, when to use simple random, systematic, stratified, cluster, multistage, convenience, purposive, quota, snowball, and volunteer sampling, and how sampling frames, eligibility rules, nonresponse, sample size, and qualitative saturation affect the final study.
For the broader relationship among research questions, designs, qualitative, quantitative, and mixed approaches, use the research design guide. For detailed data-collection tools such as surveys, interviews, focus groups, and observation, use the research methods guide.
What Is Sampling in Research?
Sampling is the process of selecting units from a larger population or body of possible evidence so the researcher can study those selected units. The units may be people, households, schools, hospitals, documents, events, organizations, neighborhoods, records, or other cases. The correct unit depends on the research question.
Sampling language is easiest to understand when the main terms are separated. A target population is the broad group to which the research question refers. The accessible or study population is the part of that group the researcher can realistically reach. A sampling frame is the operational list or mechanism used to identify eligible units for selection. The sample is the set actually selected or recruited, and the analytic sample is the set that remains after eligibility, participation, missing-data, or quality rules are applied.
| Term | What it means | Example |
| Target population | The full population about which the study seeks knowledge. | All first-year students enrolled at public universities in a country. |
| Accessible population | The portion of the target population the research team can reach. | First-year students at six participating public universities. |
| Sampling frame | A list or operational source from which units can be selected. | Current enrollment rosters from the six universities. |
| Sample | The units selected or recruited for the study. | A selected group of 900 students. |
| Analytic sample | The observations included in the final analysis after study rules are applied. | 842 students with eligible, usable responses. |
| Practical rule
Always write down the target population, sampling unit, frame or recruitment source, eligibility criteria, and selection method before discussing sample size. This prevents a large but poorly defined sample from looking more rigorous than it is. |
Why the Sampling Method Matters
The sampling method helps determine the strength and type of inference a researcher can make. In probability sampling, units on a defined frame are selected using a mechanism with known selection probabilities. This creates a design-based foundation for estimating population quantities and sampling uncertainty, although coverage problems, nonresponse, measurement error, and implementation failures can still introduce bias.
In non-probability sampling, the chance that each eligible unit will be selected is not known. Participants may be chosen because they are easy to reach, meet a particular criterion, belong to a hard-to-reach network, represent a needed subgroup, or volunteer. These approaches can be entirely appropriate for exploratory, qualitative, pilot, specialized, or access-constrained studies, but they do not automatically support the same population-level statistical inference as a well-executed probability design.
Sampling also shapes who is visible in the evidence. A frame that omits people without stable internet access, a convenience sample drawn only from one clinic shift, or a focus-group recruitment strategy that excludes non-English speakers can systematically narrow the perspectives observed. Good sampling therefore requires thinking about coverage and inclusion, not only randomization.
Probability vs. Non-Probability Sampling
The broadest way to organize sampling methods in research is to separate probability from non-probability approaches. The key difference is whether selection probabilities are known through the sampling design, not whether the researcher uses a computer, whether the sample is large, or whether the participants seem diverse.
| Feature | Probability sampling | Non-probability sampling |
| Selection mechanism | Uses a probability-based selection process from a defined frame or sampling mechanism. | Uses availability, judgment, quotas, networks, volunteering, or another non-random mechanism. |
| Selection probability | Known or derivable under the design; it need not be equal for every unit. | Unknown for at least some eligible units. |
| Typical purpose | Population estimation, prevalence, comparisons, survey inference, representative coverage goals. | Depth, exploration, specialized cases, hard-to-reach populations, pilots, qualitative inquiry, rapid or constrained studies. |
| Common methods | Simple random, systematic, stratified, cluster, multistage, probability proportional to size. | Convenience, purposive, quota, snowball, volunteer/self-selection. |
| Main caution | Random selection does not remove coverage error, nonresponse, measurement error, or poor implementation. | A large or diverse-looking sample is not automatically representative of the target population. |
| Analysis implication | Estimation should reflect the actual sample design, including weights, strata, or clusters when relevant. | Population inference often requires stronger assumptions or specialized adjustment/modeling and transparent limitations. |
Probability Sampling Methods
Probability sampling is most useful when the study needs defensible estimates about a defined population and the researcher has, or can construct, a workable mechanism for probability-based selection. The exact design should reflect the structure of the population, the available frame, cost, geography, subgroup needs, and analysis plan.
1. Simple Random Sampling
In a simple random sample, every eligible unit on the frame has an equal chance of selection and every sample of the specified size is selected through a random mechanism. The method is conceptually straightforward and can work well when a complete, accurate frame exists and the population is not prohibitively dispersed.
Example: A university has a verified list of 8,000 final-year students and needs 500 for a graduation-experience survey. The researcher assigns each student an identifier and uses a random process to select 500 without replacement. The strength of the design depends on the roster covering the intended population and on what happens after selection, including contact and nonresponse.
2. Systematic Sampling
Systematic sampling selects units at a regular interval after a random start. If 10,000 records are ordered on a suitable frame and the study needs about 1,000, the researcher may choose a random starting position among the first 10 records and then select every 10th record. The method can be easier to implement than drawing thousands of separate random numbers.
Systematic sampling needs attention to how the list is ordered. If the ordering contains a repeating pattern that aligns with the sampling interval, the resulting sample can be distorted. Researchers should understand the frame before treating “every kth case” as automatically random.
3. Stratified Random Sampling
Stratified sampling first divides the population into non-overlapping strata based on a characteristic relevant to the study, then selects a probability sample within each stratum. Common strata include region, year of study, facility type, age group, or another characteristic known before selection. The design is valuable when the study needs reliable representation or estimates for important subgroups.
Example: A nursing workforce study needs estimates for urban, peri-urban, and rural hospitals. Instead of drawing one unrestricted sample that might include too few rural nurses, the researcher stratifies the frame by hospital location and randomly samples within all three strata. If unequal sampling fractions are used, the analysis may require weights that reflect those probabilities.
4. Cluster Sampling
Cluster sampling divides the population into naturally occurring groups such as schools, villages, clinics, census areas, or work sites, then selects clusters rather than directly sampling individuals across the entire population. The researcher may study all eligible units in selected clusters or take an additional sample within each selected cluster.
Cluster sampling is often chosen for logistical efficiency when the population is geographically dispersed. Its analysis must account for the fact that units inside the same cluster can be more similar to one another than units selected independently across the full population.
5. Multistage Sampling
Multistage sampling selects units in two or more probability-based stages. A national education study might first sample regions, then schools within selected regions, then classrooms within schools, then students within classrooms. Multistage designs are common when no single frame lists every final sampling unit or when direct nationwide sampling would be too costly.
Because the probability of inclusion is built across stages, the study team must preserve enough information about each stage to calculate or apply appropriate selection weights and analysis procedures.
6. Probability Proportional to Size Sampling
In some cluster or multistage designs, clusters vary greatly in size. Probability proportional to size (PPS) gives larger clusters a greater probability of selection, usually based on a known size measure such as enrollment, household count, or facility volume. PPS can help create efficient designs, but it is more technical than simple random selection and should be planned together with the intended estimator and weights.
Stratified vs. Cluster Sampling: The Difference Students Often Miss
Stratified and cluster sampling both divide a population into groups, but they use those groups differently. In stratified sampling, the researcher usually wants representation from every stratum and samples within each one. In cluster sampling, the researcher selects only some clusters, often for efficiency, and studies units within the selected clusters.
| Question | Stratified sampling | Cluster sampling |
| Why form groups? | To ensure or improve representation/precision for meaningful subgroups. | To make selection and data collection more practical or efficient. |
| Which groups enter the sample? | Usually every stratum contributes sampled units. | Only selected clusters contribute units. |
| Ideal group structure | Units within a stratum are similar on the stratifying variable; strata differ from one another. | Clusters often resemble mini-populations, though real designs vary. |
| Example | Sample students separately from first, second, third, and fourth year. | Randomly select 12 schools, then sample students within those schools. |
| Analysis issue | May require stratum information and weights. | Must account for within-cluster dependence and the multistage selection structure when applicable. |
| Memory aid
Stratified sampling usually samples from each group. Cluster sampling usually samples some groups. That shortcut is not the full statistical definition, but it prevents the most common introductory mistake. |
Non-Probability Sampling Methods
Non-probability sampling is appropriate when probability selection is not feasible or when the research purpose values access, depth, diversity of experience, specialized knowledge, or theoretical relevance more than design-based population estimation. The limitation is not that every non-probability study is weak; the limitation is that the claims must match how participants were selected.
1. Convenience Sampling
Convenience sampling recruits units that are readily available. Examples include surveying students leaving one lecture hall, using patients who attend a clinic during the researcher’s placement period, or analyzing records from the one organization that granted access. It is fast and practical, which can make it useful for pilots or classroom projects, but availability can be strongly related to the outcome being studied.
2. Purposive Sampling
Purposive sampling deliberately selects people or cases because they have characteristics, experiences, knowledge, or perspectives relevant to the research question. It is central to many qualitative studies. A researcher studying the implementation of a new hospital policy might deliberately include nurse managers, frontline nurses, implementation leads, and staff from units with different adoption experiences.
Purposive sampling is a family of strategies rather than a single formula. Researchers may use criterion sampling, maximum-variation sampling, typical-case sampling, extreme-case sampling, expert or key-informant sampling, or theoretically driven sampling depending on the study purpose.
3. Quota Sampling
Quota sampling sets target numbers for selected characteristics, such as age group, gender, region, or role, but fills those quotas through non-random recruitment. A study might aim for 100 respondents from each of four age bands but recruit whichever eligible participants are accessible until each target is full. Quotas can improve visible balance across chosen characteristics, but they do not create known selection probabilities within the quotas.
4. Snowball or Chain-Referral Sampling
Snowball sampling begins with initial participants who help identify or refer other eligible participants. It can be useful when the population is difficult to locate through a conventional frame, such as members of informal occupational networks or people with uncommon experiences. Because recruitment travels through social connections, people with larger or more visible networks may be easier to reach, and closely connected participants may share similar characteristics.
More advanced network-based designs, such as respondent-driven sampling, add formal procedures and estimators and should not be treated as identical to an informal snowball sample.
5. Volunteer or Self-Selection Sampling
Volunteer sampling occurs when eligible people decide for themselves whether to enter after seeing an invitation, link, advertisement, or open call. Online polls and open survey links are common examples. Volunteers may differ systematically from non-volunteers in motivation, topic interest, available time, digital access, or experience, so a high response count does not by itself remove self-selection bias.
Sampling in Qualitative Research
Qualitative research typically uses a different sampling logic from probability surveys. The objective is often to obtain information-rich cases that illuminate how people interpret, experience, or navigate a phenomenon. CDC guidance notes that qualitative samples are commonly small and purposively selected, with participants chosen for relevant experiences, characteristics, knowledge, or community position.
A qualitative sample may therefore be designed for depth and variation rather than numerical representativeness. A study of discharge communication might intentionally recruit patients with different ages, language needs, diagnoses, discharge destinations, and readmission experiences so the analysis can examine how the process varies across cases.
Sample size in qualitative work should be justified by the design, population, information needs, and analytic strategy. Saturation is often discussed as the point at which additional data collection yields little or no substantively new insight, but it should not be treated as a magical universal number. Researchers should explain what kind of saturation or sufficiency they sought and how they judged it.
Sampling Frames, Coverage, and Eligibility
A probability design can be technically correct and still fail to cover the intended population if the sampling frame is incomplete. A voter list excludes unregistered adults. An employee directory may omit contractors. A clinic register may omit people who never reached care. An online panel may exclude people without adequate digital access. Coverage problems matter when the omitted group differs in ways relevant to the research question.
Eligibility criteria should be defined before recruitment begins. Inclusion criteria specify who or what can enter the study; exclusion criteria identify cases that appear relevant but should not be included for defensible methodological, ethical, or safety reasons. Criteria should come from the research question and design, not from a desire to remove inconvenient cases after results are visible.
AAPOR transparency standards emphasize reporting whether a sample is probability-based or non-probability-based, describing the frame or recruitment source, explaining coverage, eligibility, oversampling, and how participants were contacted or selected. That reporting principle is useful well beyond public-opinion surveys because readers need to see how the observed sample arose.
How to Choose a Sampling Method Step by Step
There is no universally best sampling method. The best choice is the one that gives the study the evidence it needs while remaining feasible, ethical, transparent, and aligned with the intended claims. Use the following workflow to choose among sampling methods in research.
- Define the target population precisely. State who or what the study is about, including setting, geography, time period, or other boundaries.
- Identify the sampling unit. Decide whether selection occurs at the level of individuals, households, schools, clinics, records, events, documents, or another unit.
- Map the accessible population and frame. List the sources from which units can actually be identified. Note who is missing from those sources.
- Decide what inference the study needs. If the goal is population estimation, probability sampling may be important. If the goal is depth, rare experience, theory development, or specialized insight, a purposive or other non-probability strategy may be more appropriate.
- Choose probability or non-probability logic. Base this decision on the research purpose, frame, feasibility, and intended claims rather than on which method sounds more advanced.
- Select the specific method. Choose simple random, systematic, stratified, cluster, multistage, convenience, purposive, quota, snowball, volunteer, or another defensible approach.
- Write eligibility and exclusion criteria. Define them before recruitment so selection remains transparent and consistent.
- Plan subgroup representation or oversampling when needed. If small but important subgroups need reliable analysis, plan for them deliberately rather than hoping enough cases appear.
- Determine sample-size needs. Use the design, expected precision, analysis, effect size, power, resources, or qualitative information-sufficiency logic appropriate to the study.
- Plan recruitment and nonresponse management. Document contact attempts, recruitment channels, refusals, attrition, replacement rules, incentives, and follow-up procedures where relevant.
- Match analysis to the sample design. Complex probability samples may require weights, strata, clusters, or design-based variance estimation. Non-probability samples require careful limitations and sometimes specialized adjustment/modeling.
- Document the final sample transparently. Report the target population, frame, method, recruitment, final sample, response or participation patterns, deviations, and limitations.
Sample Size Is Not the Same as Sampling Quality
Sample size affects precision and analytic capability, but it cannot repair every sampling problem. If a survey recruits 100,000 volunteers from a platform that systematically excludes or underreaches important parts of the target population, the result can still be biased. A smaller probability sample drawn from a strong frame may provide a better foundation for population estimates.
For quantitative studies, sample-size planning may depend on the outcome, desired precision, expected variability or event rate, subgroup analyses, design effect, power, effect size, number of predictors, attrition, and available resources. Complex cluster designs often need larger nominal samples than simple random designs to achieve similar precision because observations within clusters can be correlated.
For qualitative studies, the relevant question is usually not statistical power. Researchers consider the specificity of the sample, complexity of the question, heterogeneity of experience, study design, quality and depth of interviews or observations, and whether additional cases continue to produce substantively new insights. A credible justification is more useful than copying a universal number from another paper.
| Do not write this
“The sample is representative because it is large.” Size alone does not establish representativeness. Explain the frame, selection method, coverage, participation, and any adjustment or weighting that supports your claim. |
Sampling Bias, Error, and Nonresponse
Sampling error is the random difference that arises because a probability study observes a sample rather than the entire population. Bias is different: it is systematic distortion. A study can have a small estimated sampling error and still be biased if the frame misses relevant groups, selected people do not respond, measures are poor, or recruitment favors certain participants.
| Problem | How it arises | Why it matters / response |
| Coverage error | The frame does not adequately include the target population. | Identify undercovered groups; improve or combine frames; limit claims if coverage cannot be repaired. |
| Selection bias | Selection mechanisms favor some eligible units over others in a way related to the outcome. | Use probability selection where needed or explain the non-probability logic and limits. |
| Nonresponse bias | Selected units do not participate and respondents differ meaningfully from nonrespondents. | Track response patterns, improve follow-up, consider weighting/adjustment where justified, and report limitations. |
| Volunteer bias | People with stronger interest, motivation, time, or experience self-select into the study. | Avoid treating open-link participation as equivalent to random selection. |
| Network bias | Chain-referral recruitment follows social ties and may overrepresent highly connected networks. | Use multiple seeds/entry points where appropriate and be cautious about population claims. |
| Attrition bias | Participants drop out over time and loss is related to exposure, outcome, or participant characteristics. | Plan retention, document loss, compare available characteristics, and address missingness analytically when appropriate. |
| Replacement bias | Researchers casually substitute easy-to-reach units for originally selected units. | Use pre-specified replacement rules or avoid replacement when it undermines the design. |
Worked Examples Across Disciplines
| Research goal | Defensible sampling approach | Why it fits |
| Estimate the prevalence of food insecurity among students at a university. | Stratified probability sample by study level or campus, drawn from current enrollment records. | Supports population estimation while ensuring smaller student groups are not missed. |
| Understand how emergency nurses experience workplace violence. | Purposive qualitative sample of nurses with relevant exposure, with variation in shift, experience, and unit. | Prioritizes information-rich experience and variation rather than prevalence estimation. |
| Compare customer satisfaction across 40 retail branches. | Cluster or multistage sample: select branches, then sample customers within selected branches. | Reduces fieldwork burden while retaining probability-based selection if stages are implemented correctly. |
| Explore the experiences of members of a difficult-to-identify informal worker network. | Purposive initial recruitment plus carefully managed snowball referrals. | A conventional frame may not exist; network recruitment can improve access, with clear limits on representativeness. |
| Ensure enough rural participants for subgroup analysis in a national survey. | Stratify by region and oversample rural strata, then use appropriate weights in population estimates. | Designs representation into selection instead of relying on chance. |
| Pilot a new questionnaire before a larger survey. | Small purposive or convenience sample that includes people with relevant characteristics and a range of likely interpretations. | The goal is instrument learning rather than population estimation, so random selection may not be necessary. |
How Sampling Connects to Research Design and Data Collection
Sampling does not stand alone. A research design guide helps determine the type of evidence needed; the sampling strategy identifies which cases can provide that evidence; and the data-collection method determines how information is gathered from those cases. Changing one part can change the others.
For example, a quantitative cross-sectional prevalence study may need a probability sample and standardized measures. A qualitative phenomenological study may use purposive sampling and in-depth interviews. A mixed methods research project may need two related samples: a broader quantitative sample followed by a purposively selected qualitative subsample chosen from survey patterns that require explanation.
The completed primary vs secondary research guide is also useful here: a project using existing records still has a sampling problem because the researcher must define which records, time periods, facilities, or cases enter the dataset. The next data collection methods in research cluster will focus on recruitment, instruments, procedures, and managing data once the sample strategy is set.
Common Sampling Mistakes
- Calling a convenience sample “random” because participants were approached in no particular order.
- Claiming a sample is representative only because it is large.
- Confusing stratified sampling with cluster sampling.
- Describing the sample size without naming the target population, frame, or recruitment source.
- Using an online link and assuming every member of the target population had an equal chance to participate.
- Ignoring the people or cases that the sampling frame cannot reach.
- Changing eligibility criteria after recruitment starts without documenting the change.
- Replacing unavailable selected participants with easier substitutes without a pre-specified rule.
- Using snowball sampling but making precise prevalence claims about the whole population.
- Applying a margin of sampling error to a non-probability sample as if it were a simple random sample.
- Using qualitative saturation as a fixed number copied from another study instead of explaining information sufficiency in the present study.
- Forgetting that complex probability samples require analysis methods consistent with strata, clusters, and weights.
Sampling Methods Checklist Before Data Collection
| Check | Question to ask |
| Population | Is the target population stated precisely enough that another researcher could identify who or what belongs in it? |
| Unit | Have I named the sampling unit and the unit of analysis? Are they the same or different? |
| Frame / access | What list, location, network, register, database, organization, or recruitment channel gives access to eligible units? |
| Coverage | Who is missing or harder to reach through that frame or channel? |
| Method | Is the sample probability-based or non-probability-based, and have I named the specific method correctly? |
| Eligibility | Are inclusion and exclusion criteria written before recruitment? |
| Subgroups | Do any important groups require stratification, quotas, purposive inclusion, or oversampling? |
| Sample size | Is the size justified using logic appropriate to the design and intended analysis? |
| Recruitment | Are contact, consent, follow-up, incentive, refusal, and replacement procedures clear? |
| Analysis | Does the planned analysis account for weights, strata, clusters, selection, or non-probability limitations where relevant? |
| Ethics | Does recruitment avoid coercion, inappropriate gatekeeping, privacy breaches, or unjust exclusion? |
| Reporting | Can I clearly report how the final sample was selected, who participated, and what limits the findings? |
Final Takeaway
Choosing among sampling methods in research is an argument about evidence, not a box-ticking step. Define the population and sampling unit, understand the frame or recruitment source, decide what type of inference the study needs, select the sampling method that fits that purpose, and plan for coverage, nonresponse, subgroup representation, ethics, sample size, and analysis.
The strongest sampling section lets a reader reconstruct how the observed cases entered the study and understand what those cases can legitimately tell us. When the research question, design, sampling strategy, data collection, and analysis all point toward the same target population and claim, the study becomes more transparent, more defensible, and easier to interpret.
Frequently Asked Questions
What are the main sampling methods in research?
They are usually grouped into probability and non-probability approaches. Common probability methods include simple random, systematic, stratified, cluster, and multistage sampling. Common non-probability methods include convenience, purposive, quota, snowball, and volunteer sampling.
Is random sampling the same as random assignment?
No. Random sampling concerns how participants or units are selected from a population. Random assignment concerns how already enrolled participants are allocated to study conditions in an experiment. A study can use one without the other.
Which sampling method is best?
There is no universal best method. The strongest choice depends on the research question, target population, available frame, intended inference, design, analysis, ethics, time, and resources.
What is the difference between stratified and cluster sampling?
Stratified sampling usually selects units from every stratum so important subgroups are represented. Cluster sampling selects some naturally occurring groups and then studies units within those selected groups, often to improve feasibility.
Can convenience sampling ever be acceptable?
Yes, especially for pilots, classroom projects, instrument testing, early exploratory work, or situations with genuine access constraints. The important requirement is to describe the recruitment honestly and limit the claims to what the sample can support.
How large should my sample be?
There is no single number. Quantitative sample-size planning depends on the design, outcome, desired precision or power, analysis, subgroup needs, expected attrition, and sometimes clustering. Qualitative studies use design- and information-based logic such as depth, variation, and saturation rather than a universal statistical formula.
Does a probability sample guarantee no bias?
No. Probability selection provides a strong basis for inference, but coverage gaps, nonresponse, measurement error, attrition, and implementation problems can still bias results.
Do qualitative studies need random sampling?
Usually not. Many qualitative studies deliberately use purposive sampling because the goal is rich understanding of relevant experiences or cases rather than estimating how common something is in the full population.
Can I use more than one sampling method?
Yes. Multistage probability designs combine methods across stages, and mixed methods projects may use different sampling strategies for quantitative and qualitative components. The relationship among the samples should be planned and reported explicitly.
