Sampling methods are of two kinds: probability sampling (every unit has a known chance of selection) and non-probability sampling (selection is based on convenience or judgment).
A. Probability sampling
- Simple random sampling: every unit has an equal chance, e.g. drawing 50 roll numbers out of 500 by lottery or random numbers. Unbiased, but needs a complete list.
- Systematic sampling: choose a random start, then every \(k\)-th unit, where \(k = N/n\). E.g. from 1000 customers, pick every 20th to get 50.
- Stratified sampling: divide the population into homogeneous groups (strata) such as gender or year of study, then take a random sample from each, often in proportion to stratum size. Gives better precision when strata differ.
- Cluster sampling: divide the population into clusters (e.g. colleges or villages), randomly select some clusters and study all units in them. Cheap for spread-out populations, but less precise.
- Multi-stage sampling: sampling in stages, e.g. select districts, then villages, then households. Used in large national surveys.
B. Non-probability sampling
- Convenience sampling: whoever is easily available, e.g. surveying friends in your class. Quick but often biased.
- Judgment (purposive) sampling: the researcher picks units thought to be typical, e.g. selecting expert teachers for an opinion survey.
- Quota sampling: fill fixed quotas for each group (e.g. 30 men, 30 women), but choice within the group is not random.
- Snowball sampling: existing participants refer others; useful for hard-to-reach groups.
Choosing a method. Probability methods allow valid statistical inference and calculation of sampling error; non-probability methods are cheaper and faster but results cannot be safely generalised to the population.