A sample in research is a smaller group selected from a larger population to represent that population in a study. Instead of collecting data from every single member of a population (which is usually impossible), researchers gather data from a sample and use statistical methods to draw conclusions about the whole group. The size of a sample is always smaller than the total population it’s drawn from.
Samples, Populations, and Why the Difference Matters
Every study starts with a target population: the entire group a researcher wants to learn something about. That population doesn’t have to be people. It could be a group of hospitals, countries, chemical compounds, animal species, or historical events. Whatever the research question applies to, that’s the population.
The sample is the slice of that population the researcher actually studies. A measurement taken from the entire population is called a parameter. A measurement taken from a sample is called a statistic. The whole point of sampling is to use that statistic to estimate the parameter as accurately as possible. If you survey 2,000 adults about their sleep habits to understand sleep patterns across the country, the national average is the parameter and your survey result is the statistic. The gap between the two is called sampling error, and keeping that gap small is one of the central challenges in research design.
Probability vs. Non-Probability Sampling
Sampling methods fall into two broad categories based on how participants are chosen.
Probability sampling gives every member of the target population an equal (or at least known) chance of being selected. This is the gold standard for generalizability because the selection process itself prevents the researcher’s assumptions or convenience from skewing the sample. The most common types include:
- Simple random sampling: Every individual has an equal chance of selection, like drawing names from a hat. It’s straightforward but can be impractical with very large or geographically spread populations.
- Stratified sampling: The population is divided into subgroups (strata) based on a shared characteristic, such as age or income level, and then random samples are drawn from each subgroup. This ensures that important subgroups are proportionally represented.
- Cluster sampling: The population is divided into naturally occurring clusters, like schools or neighborhoods, and entire clusters are randomly selected for study. This is cheaper and more practical when studying large, dispersed populations.
- Systematic sampling: Researchers select every nth individual from a list, such as every 10th patient on a hospital registry. It’s simpler to execute than pure random sampling and works well when the list has no hidden pattern.
Non-probability sampling does not guarantee every member of the population an equal chance of being included. It’s more practical and less expensive, which is why it’s extremely common, especially in clinical research. But it carries a higher risk of producing a sample that doesn’t accurately reflect the population. Common types include:
- Convenience sampling: Researchers recruit whoever is available and accessible. It’s the most widely used method in clinical research because it’s fast and affordable, but it can easily introduce bias.
- Purposive sampling: Researchers deliberately choose participants who meet specific criteria relevant to the study, such as people with a rare disease or experts in a particular field.
- Snowball sampling: Existing participants recruit others from their networks. This is useful for reaching hard-to-find populations, like undocumented workers or people with stigmatized conditions.
- Quota sampling: Researchers set a target number for certain subgroups (for example, 50 men and 50 women) and then use non-random methods to fill each quota.
How Researchers Decide on Sample Size
Choosing the right number of participants isn’t guesswork. Researchers typically perform a calculation called a power analysis before the study begins, weighing four key factors:
The first is effect size, which is the minimum meaningful difference the study is trying to detect. A study looking for a tiny difference between two treatments needs more participants than one looking for a dramatic difference. The second is the significance level (often set at 5%), which represents the maximum acceptable risk of a false positive, concluding something is real when it isn’t. The third is statistical power (typically set at 80% or higher), which is the probability of correctly detecting a real effect. The fourth is variability: populations with more diverse characteristics require larger samples to produce reliable estimates.
Larger samples generally produce more precise results. As sample size grows, the margin of error shrinks and confidence intervals narrow, meaning researchers can be more certain their findings reflect reality rather than random chance.
Sampling in Qualitative Research
Qualitative research, which focuses on understanding experiences and meanings rather than measuring numerical outcomes, approaches sampling differently. Instead of calculating a target number upfront, many qualitative researchers use a concept called saturation. This means continuing to collect data, typically through interviews or observations, until new participants stop revealing new information. When the same themes and patterns keep appearing, the sample is considered sufficient.
This idea originated in grounded theory methodology, where sampling, data collection, and analysis happen simultaneously rather than in separate stages. The researcher recruits new participants specifically to test and refine emerging ideas, seeking out people who might challenge or extend patterns found so far. Sampling stops when the theoretical categories being developed are fully fleshed out and additional data no longer adds new dimensions. Qualitative samples are therefore typically much smaller than quantitative ones, often ranging from a dozen to a few dozen participants, but the depth of data from each participant is far greater.
What Makes a Sample Representative
A sample is representative when the findings it produces can be generalized to the target population. There are two ways this works. The stronger form is quantitative generalizability: the distributions of key characteristics (age, sex, health status, income, or whatever matters for the study) in the sample closely match those in the population. Random sampling achieves this by design. Even when the overall distributions differ, a sample can still be representative within specific subgroups, so long as those subgroups are measured and accounted for in the analysis.
The weaker form is qualitative generalizability: even if the sample’s numbers don’t perfectly mirror the population, the underlying biological processes or mechanisms being studied are likely the same. For example, a drug trial conducted in one country might still be informative for patients elsewhere if the disease works the same way in both groups. This kind of generalization requires more caution and relies on assumptions that should be stated clearly.
Two things are always required for either type. First, the target population must be clearly defined. You can’t assess whether a sample represents a population if you haven’t spelled out what that population is. Second, researchers need to be transparent about their assumptions for why findings from the sample would apply more broadly.
Selection Bias and Sampling Error
These two problems are easy to confuse but fundamentally different. Sampling error is the natural, unavoidable gap between your sample statistic and the true population value. It happens even with perfect sampling because you’re studying a subset, not the whole. Larger samples reduce it; smaller samples increase it. It’s random and predictable.
Selection bias is systematic, not random, and it happens when the people (or items) studied aren’t representative of the target population. Consider a researcher trying to estimate how many city residents drink heavily by mailing questionnaires to patients registered with local doctors. People not registered with a doctor are excluded entirely, and they may have very different drinking patterns. On top of that, people who don’t bother returning the questionnaire might drink more (or less) than those who do. Both of these problems introduce selection bias, tilting the results in one direction regardless of how large the sample is. Unlike sampling error, you can’t fix selection bias simply by adding more participants. It requires better study design.
How Samples Are Reported in Published Research
Published studies are expected to describe their samples in enough detail that readers can judge the findings for themselves. Major reporting guidelines used across medical and scientific journals require authors to explain how participants were selected, how many were included, what the eligibility criteria were, and how the sample relates to the broader population. For clinical trials, this includes describing the randomization process. For observational studies like surveys or cohort studies, it means detailing recruitment methods and potential sources of bias. This transparency allows readers and reviewers to assess both internal validity (whether the study’s conclusions are sound for the sample studied) and external validity (whether those conclusions apply beyond the sample to the wider population).