Probability and Non-Probability Sampling Techniques
Sampling selects units from a population so evidence can be collected efficiently. The method determines selection probabilities, representativeness, feasible inference and vulnerability to bias.
Learning objectives
- Differentiate probability from non-probability sampling.
- Apply simple random, systematic, stratified, cluster and multistage procedures.
- Explain consecutive, convenience, purposive, quota and snowball sampling.
- Select and justify a method for realistic health scenarios.
Probability sampling
| Method | Procedure | Best use |
|---|---|---|
| Simple random | Select unique numbers from a complete frame | Relatively small, well-listed population |
| Systematic | Random start, then every kth unit | Ordered clinic or household list without harmful periodicity |
| Stratified | Sample separately within important subgroups | Ensuring sex, cadre or facility representation |
| Cluster | Select natural groups, then study all or a sample within them | Geographically dispersed populations |
| Multistage | Sample through successive levels | National or district community surveys |
Systematic sampling example
A clinic expects 1,200 eligible patients and needs 200. The interval k = 1,200/200 = 6. Randomly choose a start from 1–6, for example 4, then select records 4, 10, 16 and so forth. Check whether list order creates periodic bias.
Stratified versus cluster sampling
Stratification
All strata are represented; sampling occurs within each. It can improve precision when strata are internally similar.
Clustering
Only selected clusters are studied. It reduces travel but observations within clusters tend to resemble one another, increasing design effect.
Non-probability sampling
- Convenience: recruit the easiest available participants; quick but highly bias-prone.
- Consecutive: include every eligible participant presenting until the sample is reached.
- Purposive: select information-rich participants with relevant characteristics.
- Quota: recruit until predefined category numbers are filled, without random selection.
- Snowball: initial participants refer others; useful for hidden populations but network-dependent.
- Theoretical sampling: qualitative recruitment evolves to develop an emerging theory.
Scenario-based choices
| Question | Defensible method | Reason |
|---|---|---|
| District household immunisation coverage | Multistage cluster sampling | No complete individual list and wide geography |
| Experiences of rare-disease caregivers | Purposive plus snowball | Small specialised network |
| All eligible surgical patients over three months | Consecutive sampling | Practical facility recruitment |
| Ensure each health-worker cadre is represented | Stratified random sampling | Explicit subgroup representation |
Sampling bias and non-response
A large sample does not repair biased selection. Document the frame, randomisation method, replacements, refusals and differences between respondents and non-respondents. Avoid replacing absent selected participants with convenient neighbours unless the protocol defines a probability-based procedure.
Common errors
- Calling consecutive clinic recruitment simple random sampling.
- Using purposive sampling and making population prevalence claims.
- Failing to describe random-number generation or sampling interval.
- Ignoring unequal selection probabilities in analysis.
- Confusing cluster selection with stratification.
Review questions
- Calculate a systematic interval for N = 900 and n = 150.
- Compare stratified and cluster sampling.
- Select methods for a hidden population and a district household survey.
- Why does sample size not eliminate selection bias?
References
- WHO. Health Research Methodology.
- Kothari CR, Garg G. Research Methodology.
Educational content only; complex survey designs require appropriate statistical analysis.
