Skip to content
All articles

Statistics

Sampling Techniques in Research: Probability and Non-Probability Methods Explained

Sampling means studying a carefully chosen part of a population to draw conclusions about the whole. Probability sampling gives every unit a known chance of selection and allows valid statistical inference.

Resora Research Team · 14 Nov 2026 · 5 min read

Sampling means studying a carefully chosen part of a population to draw conclusions about the whole. Probability sampling gives every unit a known chance of selection and allows valid statistical inference. Its main methods are simple random, systematic, stratified, cluster and multistage sampling. Non-probability sampling (convenience, purposive, quota, snowball, consecutive) is quicker and cheaper but cannot be generalised with confidence, so it suits pilot, qualitative or hard-to-reach-population studies. Choose the method by your objective, whether a sampling frame exists, how spread out the population is, and your budget. A good sampling method reduces bias; a good sample size reduces random error. You need both.

Why sample, and key terms

Studying everyone (a census) is usually too costly, too slow, or impossible. A well-chosen sample gives accurate answers at a fraction of the cost.

TermMeaning
Target populationEveryone you want your results to apply to
Study populationThe accessible part of the target population
Sampling frameThe list from which you draw the sample (voter list, ration cards, hospital register)
Sampling unitThe unit selected: an individual, a household, a village
Sampling errorRandom difference between sample and population; shrinks with larger samples
Non-sampling errorBias from poor design, non-response or measurement; does not shrink with larger samples

A frame that is out of date or incomplete is a common source of bias: if the list misses migrant families, so will your sample.

Probability sampling methods

Every unit has a known, non-zero chance of selection. Only these methods allow results to be generalised with confidence intervals.

1. Simple random sampling

Every unit has an equal chance of being picked, using a random number table or computer-generated numbers.

  • Example: select 100 of 1,200 women on an antenatal register by random numbers.
  • Good: unbiased and simple to analyse.
  • Limits: needs a complete frame; can scatter the sample widely and miss small subgroups.

2. Systematic random sampling

Pick a random start, then take every kth unit, where k = N / n (the sampling interval).

  • Example: N = 1,200 and n = 100 give k = 12. Choose a random start between 1 and 12 (say 7), then pick units 7, 19, 31, 43, and so on.
  • Good: easy to carry out in the field or from registers and OPD lists.
  • Limits: bias if the list has a hidden pattern that matches the interval, for example every 7th day being a market day.

3. Stratified random sampling

Divide the population into homogeneous groups (strata) such as age group, sex or urban/rural, then take a random sample within each stratum.

  • Example: a survey of 800 adults split into rural and urban strata, sampled in proportion to each group's size.
  • Good: ensures every subgroup is represented; improves precision when strata differ.
  • Limits: needs information about the stratifying variable for every unit.

4. Cluster sampling

The population is divided into natural groups (clusters) such as villages, wards or schools. Some clusters are chosen at random and all units in them are studied.

  • Example: randomly choose 30 villages and survey every eligible household in them.
  • Good: no complete list of individuals is needed; cheap when the population is spread out.
  • Limits: people in a cluster resemble each other, so precision is lower. You need a larger sample, multiplied by the design effect (usually 1.5–2).
  • The WHO 30 × 7 cluster survey for immunisation coverage (30 clusters, 7 children each) is the classic public health example.

5. Multistage sampling

Sampling in successive stages, each with its own random selection.

  • Example: randomly choose 4 districts, then 5 blocks in each, then 10 villages in each block, then 20 households in each village.
  • Good: practical for large national surveys such as NFHS.
  • Limits: complex design and analysis; errors can add up at each stage.

Non-probability sampling methods

Selection is not random, so the chance of being chosen is unknown. These methods cannot support statistical generalisation, but they are useful when a frame is unavailable or when depth matters more than representativeness.

MethodHow it worksTypical use
ConvenienceWhoever is easily availablePilot studies, quick surveys of OPD attendees
ConsecutiveEvery eligible patient who attends until the sample size is reachedHospital-based clinical studies (the most common choice in theses)
Purposive (judgement)The researcher deliberately picks information-rich casesQualitative research, key informant interviews
QuotaFixed numbers from each category, filled by convenienceOpinion polls, market research
SnowballParticipants refer others like themHidden populations: people who inject drugs, sex workers, migrants

Consecutive sampling in a hospital is practical and reduces some selection bias compared with convenience sampling, but findings still apply only to patients like yours, not to the general population.

Comparison at a glance

MethodNeeds a complete frameCostRepresentativenessBest for
Simple randomYesMediumHighSmall, listed populations
SystematicYes (a list)LowHigh (if no pattern)Registers, OPD lists
StratifiedYes, plus strata dataMediumVery highSubgroup comparison
ClusterOnly a list of clustersLowModerate (design effect)Large, spread-out populations
MultistageList at each stageLowModerateNational surveys
Non-probabilityNoLowestUnknownPilot, qualitative, hidden populations

How to choose

  1. Do you need to generalise to a population? If yes, use probability sampling.
  2. Do you have a list of all units? If yes, simple random, systematic or stratified. If not, use cluster or multistage.
  3. Are there important subgroups? Use stratified sampling.
  4. Is the population scattered over a large area? Use cluster or multistage.
  5. Are you studying a hidden or small group, or exploring a topic? Use purposive or snowball sampling.

Common mistakes

  • Calling a convenience sample "random".
  • Using a sampling frame that is outdated or misses groups.
  • Forgetting the design effect when using cluster sampling.
  • Replacing a missing household with the nearest convenient one instead of following the replacement rule.
  • Believing a very large sample fixes a biased method. Size does not remove bias.
  • Not describing the sampling method clearly enough for another researcher to repeat it.

Keywords

sampling techniques, types of sampling, probability sampling, non-probability sampling, simple random sampling, systematic sampling, stratified sampling, cluster sampling, multistage sampling, convenience sampling, purposive sampling, snowball sampling, sampling frame, sampling error, sampling methods for thesis

References

  • Park K. Park's Textbook of Preventive and Social Medicine — chapter on sampling and biostatistics.
  • Kirkwood BR, Sterne JAC. Essential Medical Statistics. 2nd ed. Blackwell Science; 2003.
  • Lwanga SK, Lemeshow S. Sample Size Determination in Health Studies. WHO; 1991.
  • Bennett S, et al. A simplified general method for cluster-sample surveys of health in developing countries. World Health Statistics Quarterly. 1991;44:98–106.
  • Mahajan BK. Methods in Biostatistics for Medical Students and Research Workers. Jaypee Brothers.

Sampling Techniques in Research: Probability and Non-Probability Methods Explained · Resora