Breaking down the question
The question carries two linked demands. The first, and the heavier, is to distinguish between probability and non-probability sampling — to draw a principled contrast between the two great families of sampling, not merely to define each. The second is a deceptively simple factual prompt — how many types of sampling designs are there — which the examiner uses to test whether you can organise the field into a clean typology rather than reciting a random list.
The command distinguish is doing real work. It asks for a comparison built on a criterion, and the governing criterion here is whether every unit in the population has a known, non-zero probability of selection. That single principle separates the two families and cascades into every other difference — the possibility of estimating sampling error, the claim to statistical generalisation, the need for a sampling frame, and the epistemological purpose each design serves.
A strong answer therefore rests on three pillars:
- A crisp statement of the defining principle that divides the two families.
- A structured enumeration of the major designs within each, so the how many types demand is met visibly.
- An evaluative close linking the choice of design to the aims of the research — generalisation versus depth.
For the wider logic of methods within which sampling sits, see the note on research methods in sociology.
How to approach it
Open with the reason sampling exists at all — the impossibility and wastefulness of studying whole populations, and the promise that a well-drawn subset can stand in for the whole. Introduce the core vocabulary early: population, sampling frame, representativeness and sampling error. This vocabulary is the axis on which your distinction turns.
Then build the contrast in two blocks. Treat probability sampling first, since it is the more rigorous family, and enumerate its principal designs — simple random, systematic, stratified, and cluster or multi-stage. Turn next to non-probability sampling and enumerate convenience, purposive, quota and snowball designs. For each design give one line on its logic, one on its strength and one on its limitation, so the answer reads as an ordered typology rather than a heap of terms.
Answer the numerical sub-question honestly. There is no single canonical count of sampling designs — texts differ — so the mature reply is to say that designs fall into two broad families, each containing roughly four principal types, giving eight commonly recognised designs, with hybrids such as multi-stage sampling multiplying the possibilities. Close by resisting the urge to crown one family as superior: the choice is problem-driven.
Model answer
Sampling is the process of selecting a subset of units — persons, households, groups or events — from a larger population so that findings from the subset may be extended, within a calculable margin, to the whole. Researchers sample because studying an entire population is usually costly, slow and often impossible, and because a carefully drawn sample can yield results that are faster, cheaper and, under tighter field control, sometimes more accurate than an unwieldy census. As Goode and Hatt put it, the sample is a miniature of the population; its scientific worth lies in representativeness.
Sampling designs divide into two families, and the line between them is a single principle: whether every unit has a known, non-zero chance of selection.
Probability sampling satisfies that principle. Because selection probabilities are known, the researcher can estimate sampling error and generalise from sample to population with stated confidence. It presupposes a sampling frame — a reasonably complete list of units. Its principal designs are:
- Simple random sampling, in which units are drawn by lottery or random numbers. It is free of selection bias and statistically firm, but demands a complete frame and can scatter the sample expensively across a wide area.
- Systematic sampling, which selects every nth unit after a random start. It is quick and easy to administer, but a hidden periodicity in the list can distort the result.
- Stratified sampling, which divides the population into homogeneous strata — by caste, income or region — and samples within each. It guarantees representation of small but important groups and improves precision, yet needs prior knowledge of the stratifying variable.
- Cluster and multi-stage sampling, which select natural groupings such as villages or wards and then sample within them. They cut cost and dispense with a full frame, but because units within a cluster tend to resemble one another, sampling error rises.
Non-probability sampling does not satisfy the principle: selection probabilities are unknown, so statistical generalisation weakens, but other virtues appear. Its principal designs are:
- Convenience sampling, which takes whoever is available — cheap and fast, useful for a pilot, but highly prone to bias.
- Purposive or judgement sampling, which selects units the researcher deems typical or information-rich. It suits qualitative and case-study work, though it leans heavily on the researcher's judgement.
- Quota sampling, which fills fixed quotas for chosen categories, mimicking stratification without randomness. Economical and common in market and opinion research, its final selection of respondents remains subjective.
- Snowball sampling, which grows through referrals and is invaluable for reaching hidden populations — sex workers, undocumented migrants, drug users — yet circulates within a network and cannot claim representativeness.
How many types, then? There is no fixed number on which all texts agree. The disciplined answer is that sampling designs fall into two families, each with about four principal designs, giving eight commonly recognised types; and because hybrids such as multi-stage and stratified-cluster combinations proliferate, the practical count is open-ended.
The two families embody a deeper divide. Probability designs serve the positivist aim of measurement and generalisation; non-probability designs serve the interpretive aim of understanding meaning in depth. In practice researchers combine them — a stratified sample for a national survey, purposive selection for follow-up interviews. The credibility of any social finding, from voting behaviour to attitudes towards untouchability, finally turns on the defensibility of its sample.
Examiner's perspective
Examiners look first for the governing criterion of distinction. A script that defines probability and non-probability sampling side by side but never names the principle — known versus unknown probability of selection — has described rather than distinguished, and answers only half the command.
The little numerical sub-question is a trap for the careless. Candidates who assert a precise figure expose a memorised rather than a reasoned grasp; the higher-scoring reply frames the field as two families of roughly four designs each, acknowledging that counts vary and hybrids exist. This shows the examiner a candidate who understands the architecture of sampling, not merely its labels.
The strongest answers close with methodological maturity — the recognition that no design is universally best, that probability and non-probability approaches reflect competing epistemologies, and that the right choice is dictated by the research problem. A concluding line on mixed-methods practice signals exactly the judgement the examiner rewards.