A sample, in the context of scientific research and statistics, is a representative subset of a population.
It's often impractical – if not impossible – to access an entire population for research or data collection. A survey involving the sleep habits of university students, for example, would be hard-pressed to collect data from all current students, and an experiment researching the effects of overpopulation in Norway rats could never include all the specimens in existence.
To get around that problem, researchers access a sample group. Characteristics of the sample should match those of the population so that the outcome of an experiment or survey conducted on a sample would be replicable if it were possible to research the whole population.
In probability-based sampling, all members of a population are equally likely to be selected, which helps ensure that the sample will be representative of the population. Researchers employ one of several random sampling methods:
Simple random sampling involves using software to randomly select subjects from the whole population.
Stratified random sampling involves creating subsets of the population based on some common factor and then randomly selecting samples from each group.
Cluster sampling involves breaking the population into separate groups, randomly selecting a subset of the groups from the population and using all members that subset.
Nonprobability-based methods include:
Convenience sampling, which involves simply collecting data from some group that is available to the researchers.
Purposive sampling, which involves defining subject criteria and then seeking out subjects that match that criteria.
Quota sampling, which involves defining some criteria for subjects that you want included in a certain percentage of samples to ensure that specific subgroups are represented.
Although nonprobability-based sampling does not ensure validity as well, it is typically simpler to conduct. In any case, however, no sampling method is infallible and researchers need to be aware of the sampling errors that can invalidate their efforts.