“Distinguish between probability and nonprobability sampling methods. How many types of sampling designs are there?” (2009)
- Sampling is the selection of a subset of a population for study, in place of a complete enumeration, and every sampling design falls into one of two broad families: probability and non-probability.
- The distinguishing logic between the two families is what this question turns on, and the count of specific types within each is best treated as an aid to memory rather than a rigid, universally agreed figure.
- Probability sampling rests on randomisation and a known chance of selection; non-probability sampling rests on the researcher’s or interviewer’s judgement.
- Large-scale quantitative surveys lean on probability designs for their statistical claims; small-scale qualitative and exploratory studies more often use non-probability designs suited to depth over representativeness.
Core Distinguishing Criteria
| Criterion | Probability Sampling | Non-probability Sampling |
|---|---|---|
| Chance of selection | Known and calculable for every unit | Unknown or non-calculable |
| Basis of selection | Randomisation | Researcher’s or interviewer’s judgement |
| Representativeness | Claims statistical representativeness of the population | Makes no claim to representativeness |
| Generalisability | Findings can be statistically generalised, with a calculable margin of error | Findings cannot be statistically generalised beyond the sample |
| Typical scale/use | Large-scale quantitative research, surveys, censuses | Small-scale qualitative, exploratory, or hard-to-reach-population research |
| Requires a sampling frame | Yes — a complete list of the population is generally needed | Not necessarily |
- The known-versus-unknown probability of selection is the technical crux: only a known, non-zero probability for every unit licenses formal statistical inference and a calculable sampling error.
- Where no sampling frame exists at all — for populations such as sex workers, substance users, or undocumented migrants — non-probability designs are often the only workable option, not merely a convenient shortcut.
How Many Types of Sampling Designs Are There
- Probability sampling is generally organised into roughly five recognised types:
- Simple random sampling — every unit has an equal and independent chance of selection, typically via a random number procedure; the baseline against which other designs are compared.
- Systematic sampling — units are selected at fixed intervals from an ordered list after a random start; efficient, but vulnerable to bias if the list has a hidden periodic pattern.
- Stratified sampling — the population is first divided into internally homogeneous strata (by region, caste category, income group, and so on) and a random sample drawn from each; it can be proportionate (strata sampled in proportion to their size in the population) or disproportionate (smaller strata deliberately oversampled to allow reliable sub-group analysis).
- Cluster sampling — the population is divided into naturally occurring clusters (villages, wards, institutions), a random sample of clusters is drawn, and all or a sub-sample of units within chosen clusters is studied; economical for geographically dispersed populations.
- Multi-stage sampling — successive rounds of sampling at different levels (for instance, districts, then villages within districts, then households within villages), combining several of the designs above in sequence.
- Non-probability sampling is likewise generally organised into roughly five recognised types:
- Convenience sampling — units are selected simply because they are easiest to access; fast and cheap, but carries the highest risk of bias.
- Purposive (judgemental) sampling — the researcher deliberately selects units believed to be information-rich or typical for the research question.
- Quota sampling — the population is divided into categories as in stratified sampling, but units within each category are then selected non-randomly, by interviewer convenience, until a fixed quota is filled.
- Snowball sampling — existing participants refer the researcher to further participants, used chiefly for hidden or hard-to-locate populations.
- Theoretical (saturation-based) sampling — associated with grounded-theory research, where further cases are selected as the analysis proceeds, guided by emerging theoretical categories, and sampling continues until additional cases add no new insight (theoretical saturation).
The Count Is Indicative, Not a Fixed Figure
- Different methodological treatments vary somewhat in exact count and terminology — some list cluster and multi-stage sampling as a single combined category, others treat proportionate and disproportionate stratified sampling as two separate types rather than sub-variants, and some add area sampling or double sampling as further probability variants.
- The number of “types” is therefore best treated as an indicative, roughly-five-and-five organising scheme for revision, not a single fixed figure to memorise as if it were settled beyond dispute.
- Real large-scale designs rarely use just one pure type: India’s National Sample Survey framework, for instance, typically compounds several probability designs together — a multi-stage design with stratification at an early stage and random or systematic selection of clusters and units at later stages — rather than relying on any single textbook type in isolation.
- The probability/non-probability distinction is best remembered through its underlying logic — known chance versus judgement, representativeness versus none — rather than through the sampling technique’s name alone.
- Each family’s internal types are tools suited to different constraints: a known population and a need for statistical inference call for probability designs, while an unlisted, hidden, or exploratory population calls for non-probability designs.
- The count of types is a teaching convenience; what actually matters for evaluating any sampling design is whether its selection procedure matches the researcher’s specific population, resources, and inferential goal.
- Contemporary large-scale Indian survey practice — spanning consumption expenditure, employment, and time-use surveys — continues to rely on compounded multi-stage probability designs precisely because no single simple design can economically cover a population as vast and diverse as India’s.