Sampling Methods and Stratification
Why Sampling Matters
A population is everyone or everything an investigation concerns. Testing every member is often impossible — too expensive, too slow, or physically impractical. Instead, researchers select a sample: a manageable subset used to draw conclusions about the whole population.
Before sampling can begin, you need a sample frame — a list of population members you can actually select from. For a school investigation, the sample frame might be the full student roll. For a national investigation, it could be the electoral register. A sample frame rarely covers every single member of the population perfectly, and any gaps are a source of bias before sampling even starts.
The word ‘population’ does not always mean everyone in a country. Its meaning depends on the context of the investigation. If a study is about staff wellbeing, the population is all employees in that office. If a study is about voting intentions, the population might be all adults registered to vote. If a study is about manufacturing defects, the population is every item produced in a factory. Always state the population that matches the specific hypothesis being tested.
Judgement and Opportunity Sampling
Judgement sampling is where the researcher deliberately picks participants they believe are typical of the wider population. It is quick and can avoid obvious outliers, but it relies entirely on the researcher’s opinion. If that opinion is wrong, the whole sample is biased in the same direction.
Opportunity sampling (also called convenience sampling) means choosing whoever happens to be available — people walking past, or friends nearby. It is cheap and fast, but the people available at a particular time or place often differ systematically from the wider population, so results can be badly skewed.
Judgement sampling relies on the researcher’s opinion about who is typical. Opportunity sampling relies on who happens to be available. Both risk bias, but for different reasons.
Cluster and Quota Sampling
Cluster sampling divides the population into naturally occurring groups — such as schools, streets, or hospital wards — and then selects entire clusters to study, rather than picking individuals one by one. It is often used when a population is spread over a wide area and visiting a full random sample would be too costly; sampling a handful of whole clusters is far more practical.
Quota sampling sets a target number for each subgroup — for example, 30 women aged 18–25 and 30 women aged 26–40 — then fills each quota by approaching whoever is convenient until the target is met. The quotas may or may not reflect the actual proportions in the population, and because selection within each quota is not random, it carries the same bias risk as opportunity sampling.
A sample chosen by judgement, opportunity, cluster, or quota methods is not automatically worthless. In early, exploratory research — trying to work out which variables are worth investigating properly — a quick, convenient sample can be useful, and speed can matter more than perfect representativeness at that stage.
However, any conclusions drawn from these methods must be clearly caveated: state that the sample may not represent the wider population, and name the specific groups most likely to be missing or over-represented.
Random Sampling
In a simple random sample, every member of the population has an equal chance of selection, and the actual choice is made by a genuinely random process. This is the method least likely to introduce systematic bias, because no particular group is favoured or excluded on purpose.
Random selection can be carried out using dice, shuffled cards, printed random number lists, or the random number function on a calculator. You may be asked in the exam to describe how to use random numbers to select a sample — for example, number every member of the sampling frame, then read numbers in order, ignoring repeats and numbers out of range, until you have enough people.
When using a random number list to select from a numbered sample frame, two practical problems come up often. First, a number might repeat — the same person could be picked twice. The standard fix is to ignore repeated numbers and move to the next one in the list.
Second, a random number might fall outside the range of the sample frame — for example, generating 847 when only 600 people are numbered. Numbers out of range are simply skipped, and the next valid number in the list is used instead.
Systematic Sampling
Systematic sampling selects every ‘k’th member of the sample frame after a random start. To select 20 people from a list of 200, calculate k = 200 ÷ 20 = 10, pick a random starting point between 1 and 10, then take every 10th person after that.
Systematic sampling is easier to apply than simple random sampling for large lists, and it spreads selections evenly across the sample frame. Its weakness is that a hidden pattern in the list can distort the sample — if a register alternates male and female entries and you sample every other name, you could end up with an all-male or all-female sample by accident.
Systematic sampling steps: (1) calculate the sampling interval k = population size ÷ sample size; (2) choose a random start between 1 and k; (3) select every k-th member after that starting point.
Stratification: A Process Before Sampling
Stratification is one of the most commonly confused ideas in this topic. It is not itself a sampling method — it is a step carried out before sampling. First, the population is divided into meaningful subgroups (strata) based on a characteristic relevant to the investigation, such as year group or gender. Random sampling is then applied separately within each stratum, in proportion to its size in the population.
Example: a college has 300 students — 120 in Year 12 and 180 in Year 13. A sample of 50 is needed. Stratifying by year group: Year 12 contributes \(\frac{120}{300} \times 50 = 20\) students, and Year 13 contributes \(\frac{180}{300} \times 50 = 30\) students. Each stratum is then sampled randomly, so the final sample reflects the year-group split of the whole college.
Stratification divides the population into subgroups before sampling begins. Random sampling is then the method actually used within each stratum — which is why the full method is called stratified random sampling. Make sure your answers show both steps.
Stratifying by More Than One Category (Higher)
At Foundation tier, stratification uses a single category, such as year group alone. At Higher tier, a population can be stratified by more than one category at once — for example, year group and gender combined. Each combination (e.g. “Year 12 male”) becomes its own stratum, and the sample size for that stratum is still calculated using the same proportion rule.
Example: of the 120 Year 12 students, 70 are female and 50 are male. In a sample of 50 from the whole college, Year 12 contributes 20 students overall (as before). Splitting that 20 by gender in the same proportion as Year 12 itself: female Year 12 students contribute \(\frac{70}{120} \times 20 \approx 11.7\), rounded to 12; male Year 12 students contribute the remaining 8. This gives a sample that reflects both year group and gender balance.
Every sampling method trades off cost and speed against the risk of bias. This is really what the specification is testing:
- Opportunity: Fast and cheap, but highly biased. Best suited to early, exploratory work.
- Judgement: Fast and avoids obvious outliers, but depends on the researcher’s assumptions being correct.
- Cluster: Practical for large, spread-out populations, but each cluster may not be typical of the whole population.
- Quota: Proportional to subgroup sizes, but not random within each quota, so it shares opportunity sampling’s bias risk.
- Random: Unbiased in principle, but needs a sample frame that genuinely covers the population.
- Systematic: Practical for large lists, but watch for hidden patterns in the sample frame.
- Stratification then random: Best when a population has important subgroups — guarantees proportional representation.