- Probability sampling gives every unit in the population a known, non-zero chance of selection, so findings can be generalized with a measurable margin of error.
- Non-probability sampling selects units by judgment, quota, or convenience; it is faster and cheaper, but the risk of sampling bias is much higher.
- Your choice depends on 5 practical factors: research objective, availability of a sampling frame, access to participants, budget, and timeline.
- Quantitative sample size comes from a statistical formula, while qualitative sample size is settled by data saturation, usually between 6 and 60 participants.
Sample vs Population: What Is the Difference?
A population is every unit you want to draw conclusions about; a sample is the smaller group you actually study. Researchers study samples because censuses are usually too slow, too costly, or physically impossible.
- A population can be finite, such as the 1,240 nurses employed by 1 hospital group, or effectively infinite, such as all future customers of a brand.
- A sample is valid only when it reflects the population on the variables that matter to the research question.
- Values calculated from a population are called parameters; values calculated from a sample are called statistics.
- The gap between a statistic and its parameter is sampling error, and it shrinks as sample size grows.
- A census is preferable only when the population is small, fully listed, and easy to reach, for example fewer than 100 employees.
| Point of comparison | Population | Sample |
| Scope | All units of interest | A selected subset of those units |
| Symbolic value | Parameter, such as mu or sigma | Statistic, such as x-bar or s |
| Cost and time | High, often prohibitive | Considerably lower |
| Measurement error | Errors accumulate across all units | Easier to control with trained staff |
| Typical use | National censuses, payroll audits | Surveys, trials, interviews, panels |
What Is a Sampling Frame?
A sampling frame is the actual list, register, or database from which you draw your sample. It is the operational version of the population: if a unit is missing from the frame, it can never be selected.
Common examples of sampling frames include:
- An electoral roll or a national identity register
- A hospital’s patient management system
- A company’s human resources database or payroll list
- A university’s enrollment records for a given semester
- A verified email panel maintained by a survey provider
A usable sampling frame should meet 4 criteria:
- Completeness: every eligible unit appears at least once.
- Accuracy: contact details and eligibility flags are current.
- Non-duplication: no unit is listed twice, which would double its selection chance.
- Relevance: the frame contains only units that meet the inclusion criteria.
When the frame and the population do not match, the result is coverage error. Undercoverage omits eligible units, such as households without landlines; overcoverage includes ineligible units, such as former employees who were never removed from the payroll list.
Types of Sampling Methods
Sampling methods are of two main types: probability and non-probability. The core difference is randomization: probability sampling uses chance and supports statistical inference, while non-probability sampling uses researcher access or judgment and supports exploration.
Probability vs Non-Probability Sampling: Key Differences
| Basis of comparison | Probability sampling | Non-probability sampling |
| Selection principle | Random selection | Judgment, quota, referral, or convenience |
| Chance of selection | Known and non-zero for every unit | Unknown and unequal |
| Sampling frame | Required | Usually not required |
| Representativeness | High when executed correctly | Not guaranteed |
| Sampling error | Can be calculated and reported | Cannot be calculated |
| Generalizability | Statistical, to the whole population | Analytical, to theory or context only |
| Cost and time | Higher | Lower |
| Typical designs | Surveys, trials, audits, censuses | Interviews, case studies, pilots, focus groups |
| Statistical tests | Fully supported | Limited and descriptive |
| Common methods | Simple random, systematic, stratified, cluster | Convenience, purposive, quota, snowball |
What Is Probability Sampling?
Probability sampling is any method in which every unit has a known, non-zero chance of being selected, decided by randomization rather than by the researcher. This is what makes inferential statistics and confidence intervals legitimate.
Use probability sampling when:
- You need results that generalize to a defined population
- You plan to run significance tests or report a margin of error
- A reliable sampling frame already exists or can be built
- The study is a survey, clinical trial, or cohort study
What Is Simple Random Sampling?
Simple random sampling selects units purely by chance, so every unit and every possible combination of units has an equal probability of selection. It is the benchmark against which other methods are judged.
- How it works: number all units in the frame, then draw the required count with a random number generator or a lottery method.
- Strengths: minimal selection bias; simple to explain and defend to reviewers.
- Limitations: needs a complete frame; small samples may miss minority subgroups entirely.
- Example: drawing 300 of 5,000 registered customers using a random number function in a spreadsheet.
What Is Systematic Sampling?
Systematic sampling selects every kth unit from an ordered list after a random start. The interval k equals the population size divided by the required sample size.
- How it works: for 1,000 units and a sample of 100, k equals 10; pick a random start between 1 and 10, then take every 10th unit.
- Strengths: fast, easy to audit, and spreads the sample evenly across the list.
- Limitations: dangerous if the list has a hidden cycle that matches the interval, for example a shift roster.
- Example: auditing every 25th invoice from a chronological ledger.
What Is Stratified Sampling?
Stratified sampling divides the population into mutually exclusive strata, such as age bands or departments, then samples randomly within each stratum. It guarantees that every subgroup appears in the final sample.
- Proportionate stratification: each stratum contributes in line with its share of the population.
- Disproportionate stratification: small but important strata are oversampled, then corrected by weighting.
- Strengths: lower sampling error than simple random sampling; supports valid subgroup comparisons.
- Limitations: requires accurate subgroup data in the frame and more complex analysis.
- Example: sampling 200 employees split across 4 grades so that each grade is represented.
What Is Cluster Sampling?
Cluster sampling divides the population into naturally occurring groups, randomly selects some of those groups, and then studies units inside them. The cluster, not the individual, is the sampling unit.
- Single-stage cluster sampling: every member of each selected cluster is studied.
- Two-stage cluster sampling: units are randomly sampled within each selected cluster.
- Strengths: cuts travel and administration costs when the population is geographically dispersed.
- Limitations: higher sampling error, because members of 1 cluster tend to resemble each other.
- Example: selecting 12 of 400 schools in a district, then surveying pupils in those 12 schools.
What Is Multistage Sampling?
Multistage sampling combines 2 or more probability methods in sequence, narrowing from large units to individual elements. National household surveys almost always use it.
- How it works: select districts at stage 1, wards at stage 2, households at stage 3, and 1 respondent per household at stage 4.
- Strengths: practical for very large populations with no single national frame.
- Limitations: error accumulates at every stage, so weighting and design effects must be calculated.
- Example: a nationwide health survey covering 28 states.
Simple Random vs Systematic vs Stratified vs Cluster Sampling: Key Differences
| Method | How units are selected | Best used when | Main limitation |
| Simple random | Every unit is drawn purely by chance from a numbered frame | The population is homogeneous and a complete list exists | Small samples can miss minority subgroups entirely |
| Systematic | Every kth unit is taken after a random start | The frame is ordered and field staff are limited | A hidden cycle matching the interval distorts the sample |
| Stratified | The population is split into subgroups, then sampled randomly within each | Subgroup comparisons are essential | Needs accurate subgroup data in the frame |
| Cluster | Whole natural groups are selected at random, then studied | The population is spread across many locations | Higher sampling error, since cluster members resemble each other |
A few points the table cannot show well:
- Precision at equal sample size: stratified is best, then simple random, then systematic, with cluster the weakest.
- Cost: cluster is cheapest for dispersed populations; stratified and simple random cost the most.
- Frame demands: simple random and systematic need a complete list of individuals; cluster only needs a list of groups.
- Analysis: cluster and disproportionate stratified designs require weighting or a design effect correction; simple random does not.
What Is Non-Probability Sampling?
Non-probability sampling selects units by convenience, judgment, quota, or referral, so the selection probability of any unit is unknown. It is common in exploratory, qualitative, and hard-to-reach population research.
Use non-probability sampling when:
- No sampling frame exists, for example with undocumented workers or rare disorders
- The goal is depth of insight rather than statistical generalization
- You are piloting an instrument or running early-stage exploratory work
- Budget or time rules out a full randomized design
What Is Convenience Sampling?
Convenience sampling recruits whoever is easiest to reach, such as passersby, students in 1 class, or visitors to 1 website. It is the fastest and least rigorous method.
- Strengths: very low cost; useful for pilot tests and instrument checks.
- Limitations: high risk of bias; results should not be presented as representative.
- Example: surveying 50 shoppers at 1 mall entrance on 1 weekday afternoon.
What Is Purposive Sampling?
Purposive sampling, also called judgmental sampling, selects participants who best fit the research question according to explicit criteria set by the researcher.
- Common variants: typical case, extreme case, critical case, maximum variation, and expert sampling.
- Strengths: produces information-rich cases; efficient for qualitative designs.
- Limitations: depends heavily on researcher judgment, which must be documented for transparency.
- Example: interviewing 15 oncology nurses with at least 10 years of experience.
What Is Quota Sampling?
Quota sampling sets targets for subgroups, then fills each target using non-random recruitment. It mimics stratified sampling in structure but not in selection logic.
- How it works: decide the quota matrix, for example 50% women and 50% men across 3 age bands, then recruit until each cell is full.
- Proportional quotas mirror population shares; non-proportional quotas guarantee a minimum count per subgroup.
- Strengths: fast, inexpensive, and keeps the sample structurally balanced.
- Limitations: interviewers may favor approachable respondents, so no margin of error can be reported.
- Example: a market research agency recruiting 400 consumers across 4 city tiers.
What Is Snowball Sampling?
Snowball sampling asks initial participants to refer other eligible people, so the sample grows through social networks. It is designed for hidden or stigmatized populations.
- Linear snowballing: each participant refers 1 more person.
- Exponential snowballing: each participant refers several people, discriminative or non-discriminative.
- Strengths: often the only workable route to hard-to-reach groups.
- Limitations: referral networks are homogeneous, so isolated individuals are systematically excluded.
- Example: recruiting 30 freelance gig workers in 1 city through peer referrals.
What Is Consecutive Sampling?
Consecutive sampling recruits every accessible unit that meets the criteria over a fixed period, until the target size is reached. It is widely used in clinical research.
- Strengths: reduces the cherry-picking seen in convenience sampling; easy to run prospectively.
- Limitations: still tied to 1 site and 1 time window, so seasonality can distort results.
- Example: enrolling all eligible patients admitted to 1 emergency department over 8 weeks.
What Is Self-Selection Sampling?
Self-selection sampling, also called volunteer sampling, lets people decide for themselves whether to take part after an open invitation. Online polls and open calls rely on it.
- Strengths: cheap, scalable, and useful when motivation to participate is itself the topic.
- Limitations: volunteers usually hold stronger opinions than non-volunteers, which inflates extreme responses.
- Example: an open survey link posted on a company intranet.
Non-Probability Sampling Methods at a Glance
| Method | How units are selected | Best used when | Main limitation |
| Convenience | Whoever is easiest to reach is recruited | Piloting a questionnaire or checking wording | Highest bias risk; results are not representative |
| Purposive | The researcher picks cases that fit explicit criteria | Depth is needed from information-rich cases | Depends entirely on researcher judgment |
| Quota | Subgroup targets are set, then filled non-randomly | Demographic balance is needed fast and on budget | Recruiters favor approachable respondents |
| Snowball | Existing participants refer other eligible people | The population is hidden, rare, or stigmatized | Referral chains stay inside 1 social network |
| Consecutive | Every accessible eligible unit is taken over a fixed period | Prospective clinical or site-based studies | Tied to 1 site and 1 time window, so seasonality intrudes |
| Self-selection | People volunteer after an open invitation | Motivation to participate is itself the topic | Volunteers hold stronger opinions than non-volunteers |
Points the table cannot capture:
- Closest probability analogue: quota mirrors stratified, and consecutive is the nearest thing to a census of an accessible window.
- Rigor ranking: consecutive and purposive are generally defensible in peer review; convenience and self-selection are the weakest.
- Reporting duty: none of these support a margin of error, so describe recruitment fully so readers can judge transferability.
- Combining methods: purposive plus snowball is a common pairing, using criteria to define who counts and referrals to find them.
What Kinds of Bias Affect Sampling?
Sampling bias is any systematic error that makes the sample unrepresentative of the population. The 9 types below account for most of the damage in applied research.
| Type of bias | What goes wrong | How to reduce it |
| Selection bias | Units are chosen in a way that favors certain characteristics | Randomize selection and pre-register the recruitment protocol |
| Undercoverage bias | Parts of the population are missing from the sampling frame | Audit and update the frame; combine multiple frames |
| Overcoverage bias | Ineligible units sit in the frame and get sampled | Apply screening questions before data collection |
| Non-response bias | People who refuse differ systematically from those who reply | Send reminders, offer incentives, and run a non-responder check |
| Self-selection bias | Volunteers hold stronger views than the average person | Use closed invitation lists rather than open links |
| Survivorship bias | Only units that lasted until measurement are studied | Track dropouts and report attrition rates |
| Convenience bias | The easiest units to reach dominate the sample | Add multiple sites, time slots, and channels |
| Network bias | Referral chains stay inside 1 social cluster | Start snowballing from several unconnected seeds |
| Healthy user bias | Participants are more engaged or healthier than non-participants | Compare sample profile against known population benchmarks |
How Do You Choose the Right Sampling Method?
Choose by working through 6 questions in order: your objective, your population, your frame, your required precision, your resources, and your ethical constraints. The first question that returns a hard limit decides the method.
- What is the objective? Generalization points to probability sampling; exploration or theory building points to non-probability sampling.
- How is the population defined? Clear inclusion and exclusion criteria must exist before any method is chosen.
- Does a sampling frame exist? Without a usable list, random selection is impossible.
- How precise must the estimate be? A tight margin of error, such as 3%, forces a larger probability sample.
- What are the budget, timeline, and field capacity? Dispersed populations often make cluster or multistage designs the only affordable route.
- What do ethics and privacy rules allow? Sensitive populations may require referral-based recruitment with informed consent.
The table below maps common scenarios to a recommended starting point.
| Your situation | Recommended method | Why it fits |
| Complete list available, homogeneous population | Simple random sampling | Lowest bias with the least design effort |
| Ordered list, limited field staff | Systematic sampling | Fast to execute and easy to supervise |
| Subgroup comparisons are essential | Stratified sampling | Guarantees each subgroup is represented |
| Population spread across many locations | Cluster or multistage sampling | Cuts travel and administration costs |
| Market research with fixed demographic targets | Quota sampling | Balances the sample quickly within budget |
| Rare, hidden, or stigmatized population | Snowball sampling | Uses trust networks to reach non-listed members |
| In-depth qualitative study | Purposive sampling | Selects information-rich cases for depth |
| Pilot test of a new questionnaire | Convenience sampling | Adequate for checking wording and flow |
How Do You Decide the Right Sample Size?
Quantitative studies calculate sample size from a formula built on confidence level, margin of error, and expected variability. Qualitative studies stop recruiting when new data stop producing new themes.
How Do You Calculate Sample Size for Quantitative Research?
Use a statistical formula or a power analysis before data collection. For a survey estimating a proportion, the required inputs are confidence level, margin of error, expected proportion, and population size.
- Confidence level: 95% is standard, giving a z value of 1.96; 99% gives 2.58.
- Margin of error: 5% is common; 3% roughly doubles the required sample.
- Expected proportion: use 0.5 when unknown, because it maximizes the required size.
- Population correction: apply the finite population correction when the population is below 10,000.
For experiments, run a power analysis targeting 80% power and a defined effect size instead. Inflate the final number by 10% to 30% to absorb non-response and dropouts.
How Do You Determine Sample Size for Qualitative Research?
Qualitative sample size is judged by data saturation, not by formula. Recruit in small waves, code as you go, and stop when 2 or 3 consecutive interviews add no new codes or themes.
- Homogeneous samples reach saturation faster than diverse ones.
- Narrow, well-defined research questions need fewer participants.
- Structured interview guides reach saturation sooner than open-ended conversations.
- Report the saturation point and the evidence for it in the methods section.
Typical published ranges by design are shown below.
| Qualitative design | Typical sample size |
| Phenomenology | 5-25 interviews |
| Interpretative phenomenological analysis | 3-10 participants |
| Grounded theory | 20-60 interviews |
| Ethnography | 1 cultural group, 30-50 interviews |
| Case study | 3-5 cases |
| Focus groups | 3-6 groups of 6-10 people each |
| Thematic analysis of interviews | 12-30 interviews |
Frequently Asked Questions
What Is the Difference Between Stratified and Cluster Sampling?
Stratified sampling samples from within every subgroup, while cluster sampling selects whole groups and ignores the rest. Stratification aims for precision; clustering aims for logistical savings.
- Strata are built to be internally similar and externally different.
- Clusters are ideally internally diverse and similar to each other.
- Stratified designs usually produce smaller sampling error than cluster designs of the same size.
Is Snowball Sampling Probability or Non-Probability Sampling?
Snowball sampling is a non-probability method, because participants are recruited through referrals and the selection probability of any individual is unknown.
- It cannot support margins of error or population-level inference.
- Respondent-driven sampling is a weighted variant that partially corrects network bias.
- Always report the number of seeds and the length of each referral chain.
What Is the Best Sampling Method for Qualitative Research?
Purposive sampling is the default choice for qualitative research, because it deliberately selects information-rich cases that can answer the research question in depth.
- Use maximum variation purposive sampling when you want contrasting perspectives.
- Add snowball sampling when the group is hidden or hard to reach.
- Use theoretical sampling in grounded theory, where each round is guided by emerging concepts.
How Many Participants Do I Need for a 95% Confidence Level?
For a large population, roughly 385 responses give a 95% confidence level with a 5% margin of error. Tightening the margin to 3% raises the requirement to about 1,067.
- Populations below 10,000 need fewer responses because of the finite population correction.
- Subgroup analysis needs the target size in each subgroup, not just overall.
- Always recruit above the calculated figure to cover incomplete responses.
What Is the Minimum Sample Size for a Thesis or Dissertation?
There is no universal minimum, but many committees expect at least 30 cases per group for quantitative work and 12 to 20 interviews for qualitative work.
- Correlational studies commonly need 100 or more respondents for stable estimates.
- Experimental designs should be justified by a power analysis rather than a rule of thumb.
Check your department handbook and your ethics approval conditions first.
How Do You Avoid Sampling Bias in Online Surveys?
Replace open public links with closed invitations to a defined frame, then track who responds and compare their profile against known population benchmarks.
- Use randomized invitation lists rather than social media broadcasts.
- Send at least 2 reminders to lift the response rate.
- Apply post-stratification weights when specific groups remain under-represented.
- Report the response rate and the completion rate together.
Can Non-Probability Sampling Be Used in Quantitative Research?
Yes, quota and convenience samples are routinely used in market research and pilot studies, but the results must be reported as indicative rather than as population estimates.
- Do not report a margin of error for a non-probability sample.
- Describe recruitment fully so readers can judge transferability.
- Consider weighting or benchmarking against a reliable reference dataset.
What Is the Difference Between a Sampling Frame and a Sampling Unit?
The sampling frame is the full list you draw from; the sampling unit is the single element you select from that list at a given stage.
- In a household survey, the frame is the address register and the unit is 1 household.
- Multistage designs use different units at each stage, such as districts, then schools, then pupils.
- A frame error affects the entire study, while a unit definition error affects comparability.
Glossary
| Term | Definition |
| Population | The complete set of people, records, or objects that a study is about. |
| Target population | The specific subset of the population that meets the study’s inclusion criteria. |
| Sample | The subset of the population that is actually observed, surveyed, or measured. |
| Sampling frame | The practical list or database from which the sample is drawn. |
| Sampling unit | The single element selected at each stage, such as 1 patient, 1 household, or 1 school. |
| Element | The individual case about which data are collected and analyzed. |
| Census | A study that collects data from every member of the population instead of a sample. |
| Sampling error | The natural difference between a sample statistic and the true population value. |
| Sampling bias | A systematic error that makes some population members more likely to be selected than others. |
| Representativeness | The degree to which the sample mirrors the population on relevant characteristics. |
| Generalizability | The extent to which results can be applied beyond the sample to the wider population. |
| Confidence level | The probability that the true value falls within the stated interval, commonly set at 95%. |
| Margin of error | The range above and below a sample estimate within which the true value is expected to lie. |
| Stratum | A mutually exclusive subgroup of the population, such as an age band or a job grade. |
| Cluster | A naturally occurring group, such as a clinic, a school, or a city block, that is sampled as a whole. |
| Response rate | The percentage of selected participants who actually provide usable data. |
| Data saturation | The point in qualitative research at which new interviews stop producing new themes. |
| Weighting | A statistical correction applied so that under-represented groups count proportionally in the results. |
Comment