Sampling Methods: Random, Stratified & Cluster Guide

Sampling methods are the procedures used to choose which members of a population end up in the sample. The method decides whether the sample fairly represents the population or silently favours some groups, and a brilliant analysis cannot rescue data collected with a biased method.

It matters because the method controls bias. A phone survey that only calls landlines misses young voters; a street survey at noon misses office workers. Choosing the right method for the population's structure is the cheapest way to buy accuracy, far cheaper than just increasing sample size.

Key Concepts

• Simple random sampling (SRS): every member has an equal chance, and every possible sample of size n is equally likely. The gold standard, often done with random numbers.
• Stratified sampling: divide the population into homogeneous groups (strata) like classes or age bands, then take a random sample from each, often in proportion to stratum size.
• Cluster sampling: divide into clusters (schools, villages), randomly select a few whole clusters, and survey everyone inside them. Cheap when the population is spread out.
• Systematic sampling: pick a random start, then take every k-th member, e.g. every 20th customer.
• Convenience and quota sampling: non-random methods, whoever is handy, or fixed quotas per group. Fast and cheap but biased; results cannot be generalised.

How It Works

Choose the method by matching it to the population. 1. SRS when you have a complete list and want maximum fairness. 2. Stratified when the population has distinct subgroups and you need each represented precisely; allocate n in proportion to stratum sizes. 3. Cluster when travel costs dominate and natural groups exist. 4. Systematic when members arrive in a stream and a list is impractical; check the list has no hidden periodicity matching k. 5. Avoid convenience sampling for any conclusion that must generalise, because its bias cannot be measured or corrected.

Worked Example

A college has 1200 students: 480 in science (40%) and 720 in arts (60%). A stratified sample of n = 120 is needed. Science share: 0.40 × 120 = 48 students; arts share: 0.60 × 120 = 72 students, each chosen randomly within their stratum. Compare with systematic sampling of the same 120 from an alphabetical roll: k = 1200/120 = 10, so pick a random start between 1 and 10, say 7, then take students 7, 17, 27, …, 1197. Both are valid probability methods; the stratified version guarantees exact subject balance.

Common Misconceptions

• Stratified and cluster are opposites: stratified samples inside every group; cluster samples a few whole groups. Mixing them up ruins the design.
• Random does not mean haphazard: grabbing whoever walks by is convenience sampling, not random sampling. True randomness needs a deliberate random mechanism.
• Systematic sampling can backfire: sampling every 7th house on streets where every 7th house is a corner shop creates a periodic bias that SRS would avoid.

Real-World Applications

• National surveys: stratified sampling by state, age and gender keeps opinion polls representative of the electorate.
• Market research in villages: cluster sampling picks whole villages to survey when reaching scattered households individually is too costly.
• Factory quality checks: systematic sampling inspects every 50th item rolling off a production line.

Sampling methods are how fairness is engineered into data: simple random for the basic guarantee, stratified when subgroups must be exact, cluster when geography is expensive, systematic when data flows in a stream.