Research Methods ·
By Reviewed against primary style manuals — see our editorial process

Sampling Methods: Probability and Non-Probability Sampling

Research rarely studies every member of a population — instead, it studies a carefully chosen subset called a sample. The way you select that sample has direct consequences for the validity and generalizability of your findings. This guide explains the full spectrum of sampling methods, from the rigorous probability approaches used in large-scale surveys to the pragmatic non-probability techniques common in qualitative research.

Population vs. sample

The population is the complete set of individuals, objects, or events that the researcher is interested in. It can be very large (all adults in a country) or narrowly defined (all patients with Type 2 diabetes treated at a specific hospital in 2025). Studying the entire population is called a census.

Because studying an entire population is usually impractical, researchers select a sample — a subset of the population — and use the data collected from that sample to make inferences about the population as a whole. The quality of those inferences depends critically on how the sample was selected.

Key terms: A parameter is a numerical characteristic of a population (e.g., the true mean income of all U.S. households). A statistic is the corresponding value calculated from a sample (e.g., the mean income of the 2,000 households surveyed). Statistics are used to estimate parameters.

Probability sampling methods

In probability sampling, every member of the population has a known, non-zero probability of being selected. This property allows researchers to calculate the margin of error and make statistically valid generalizations to the population.

Simple random sampling

Every member of the population has an equal probability of selection. Achieved by assigning numbers to all population members and drawing numbers randomly (using a random number generator or lottery). This is the simplest and most defensible method, but requires a complete sampling frame (a list of all population members), which is often unavailable.

Example

A university wants to survey 300 of its 6,000 undergraduate students. It obtains the full student roster, assigns each a number, and uses random number software to select 300.

Stratified random sampling

The population is divided into non-overlapping subgroups (strata) based on a characteristic relevant to the study (e.g., gender, age group, income bracket). A random sample is then drawn from each stratum. This ensures that important subgroups are adequately represented and typically produces more precise estimates than simple random sampling for the same total sample size.

Example

A national health survey stratifies the population by age group (18–34, 35–54, 55+) and geographic region (urban, suburban, rural), then randomly samples within each stratum to ensure all subgroups are represented.

Cluster sampling

The population is divided into clusters (e.g., schools, neighborhoods, hospitals), a random sample of clusters is selected, and all (or a random subset of) members within selected clusters are studied. Cluster sampling is cost-effective when the population is geographically dispersed, but it produces higher sampling error than simple random sampling if clusters are internally homogeneous.

Example

A researcher studying classroom practices selects 30 schools randomly from a district (clusters) and surveys all teachers in those schools rather than sampling individual teachers from across every school.

Systematic sampling

Participants are selected at regular intervals from an ordered list. A random starting point is chosen, and then every kth person is selected (where k = population size ÷ desired sample size). Systematic sampling is straightforward to implement but can introduce bias if there is a periodic pattern in the list that coincides with the sampling interval.

Example

From a list of 10,000 customers, a researcher needs 500 participants. k = 10,000 ÷ 500 = 20. A random number between 1 and 20 is drawn (say, 7), and then customers 7, 27, 47, 67… are selected.

Non-probability sampling methods

In non-probability sampling, not every member of the population has a known or equal chance of selection. These methods are faster and cheaper than probability sampling but do not allow statistical generalization to the population. They are widely used in qualitative research, pilot studies, and situations where a sampling frame is unavailable.

Convenience sampling

Participants are selected based on their availability and willingness to take part — often students in a researcher's own class, volunteers who respond to a posted advertisement, or users of a particular website. Convenience samples are easy to obtain but are highly susceptible to selection bias.

Example

A psychology researcher recruits participants from the university's introductory psychology course, offering course credit as an incentive.

Purposive (judgmental) sampling

Participants are selected deliberately based on specific characteristics that make them relevant to the research question. This is the dominant sampling strategy in qualitative research, where the goal is depth of understanding rather than statistical representation.

Example

A researcher studying expert knowledge of climate change purposively selects 15 climate scientists with at least 10 years of field experience and publications in leading journals.

Snowball sampling

An initial group of participants recruits further participants from their networks, growing the sample like a rolling snowball. This is particularly valuable for reaching hidden, marginalized, or hard-to-access populations (e.g., undocumented immigrants, illicit drug users, people with rare conditions).

Example

A sociologist studying underground cryptocurrency traders recruits three initial contacts, who each refer two more participants, who in turn refer others.

Quota sampling

The researcher specifies quotas for subgroups (e.g., 50 men and 50 women, or 40% from urban areas) and then fills those quotas using convenience or purposive selection. Quota sampling resembles stratified sampling but uses non-random selection within strata, so it cannot support probability-based inferences.

Comparing the methods

Method Type Best used when Main limitation
Simple random Probability Complete sampling frame available; homogeneous population Requires full population list; may miss subgroups
Stratified random Probability Key subgroups must be represented; heterogeneous population Requires population data on stratifying variable
Cluster Probability Population geographically dispersed; large scale Higher sampling error than simple random
Systematic Probability Ordered list available; large population Periodic patterns can bias selection
Convenience Non-probability Pilot studies; exploratory research; limited resources High risk of selection bias; not generalizable
Purposive Non-probability Qualitative studies; expert or information-rich cases needed Researcher bias in selection; not statistically representative
Snowball Non-probability Hidden or hard-to-reach populations Network bias; overrepresents certain social clusters

Sampling error and bias

Sampling error is the natural, expected difference between a sample statistic and the true population parameter, arising simply from chance variation in who is selected. Sampling error decreases as sample size increases and can be quantified with a margin of error and confidence interval in probability sampling.

Sampling bias is a systematic (non-random) error that causes the sample to differ from the population in a consistent direction. Unlike sampling error, increasing sample size does not reduce bias — it only amplifies it. Common sources of bias include:

  • Non-response bias: People who choose not to participate differ systematically from those who do.
  • Undercoverage: Some segments of the population have no chance of being selected (e.g., people without internet access in an online survey).
  • Voluntary response bias: People with strong opinions are more likely to volunteer, skewing results toward extreme views.

Determining sample size

For quantitative research, sample size is typically determined through a power analysis before data collection. The four inputs are: (1) the desired significance level (usually α = 0.05), (2) the desired statistical power (usually 0.80 or 0.90), (3) the expected effect size, and (4) the type of statistical test to be used. Larger expected effect sizes require smaller samples; smaller effects require larger ones.

For qualitative research, sample size is guided by the concept of theoretical saturation — data collection continues until new interviews or observations no longer add new themes or categories. In practice, qualitative samples of 12–30 participants are common for interview-based studies, though this varies by design.

Rule of thumb for surveys: For a simple proportion (e.g., the percentage of voters who support a policy), a margin of error of ±3% at 95% confidence requires approximately 1,067 respondents regardless of population size — assuming simple random sampling. This is why national opinion polls typically use samples of around 1,000.

Researching sampling methods and need academic sources to back your methodology section? CiteGenie finds peer-reviewed papers instantly.

Find Sources for Your Research