Topics

Sampling methods

Why sampling decides everything

A population is everyone the research is about — all students at the school, all customers of the café, all households in the suburb.
A sample is the part of that population you actually collect data from, because surveying everyone is impossible.
The sample is the single biggest threat to validity. If the sample does not represent the population, the findings describe the sample and nothing else — however many people you asked and however good your questions were.

Random sampling

Every member of the population has an equal chance of being selected — for example numbering the whole school roll and using a random number generator.
Advantages: free of researcher bias; statistically the most defensible.
Disadvantages: you need a full list of the population; the people selected may be hard to reach; a small random sample can still miss a group by chance.

Stratified sampling

The population is divided into strata (groups that matter — year level, age band, department), and people are selected from each stratum in proportion to its share of the population.
If Year 9 is 25% of the roll, Year 9 is 25% of the sample.
Advantages: guarantees every relevant group is represented in the right proportion; the most accurate of the practical methods.
Disadvantages: you need to know the population's composition beforehand — which is exactly what secondary data provides; more work to organise.

Quota sampling

The researcher decides how many people to survey from each category — 20 Year 9s, 20 Year 10s — and then surveys whoever they can find until each quota is filled.
Advantages: fast, cheap, and no population list is needed; ensures each group appears.
Disadvantages: not random, because the researcher chooses who to approach, so bias creeps in — approachable people are over-represented.

Cluster sampling

The population is divided into clusters (whole classes, whole suburbs, whole stores), a number of clusters are selected, and everyone inside the selected clusters is surveyed.
Advantages: very efficient when the population is spread out; no full list of individuals needed.
Disadvantages: people within a cluster tend to be similar, so the sample is less varied than its size suggests — three classes of Year 12 students are not a cross-section of a school.

Larger samples reduce the effect of unusual individual answers, but a large biased sample is worse than a small unbiased one — it is confidently wrong.
The standard asks for sufficient primary data of adequate quantity and quality. For a school-based project that usually means tens rather than hundreds, but it must be enough that one unusual respondent cannot move your conclusion.
Report the response rate, not just the number returned: 40 responses from 50 handed out is a much stronger result than 40 from 400.

Convenience bias — surveying whoever is nearby. Surveying at the canteen queue at lunchtime samples people who already buy lunch, which will overstate demand for a lunch product.
Self-selection bias — only people who care enough to respond do so, and they are usually the most enthusiastic or the most annoyed.
Time and place bias — a Friday afternoon sample is different from a Monday morning one.

A population is everyone the research is about — all students at the school, all customers of the café, all households in the suburb.
A sample is the part of that population you actually collect data from, because surveying everyone is impossible.
The sample is the single biggest threat to validity. If the sample does not represent the population, the findings describe the sample and nothing else — however many people you asked and however good your questions were.

Random sampling

Every member of the population has an equal chance of being selected — for example numbering the whole school roll and using a random number generator.
Advantages: free of researcher bias; statistically the most defensible.
Disadvantages: you need a full list of the population; the people selected may be hard to reach; a small random sample can still miss a group by chance.

Stratified sampling

The population is divided into strata (groups that matter — year level, age band, department), and people are selected from each stratum in proportion to its share of the population.
If Year 9 is 25% of the roll, Year 9 is 25% of the sample.
Advantages: guarantees every relevant group is represented in the right proportion; the most accurate of the practical methods.
Disadvantages: you need to know the population's composition beforehand — which is exactly what secondary data provides; more work to organise.

Quota sampling

The researcher decides how many people to survey from each category — 20 Year 9s, 20 Year 10s — and then surveys whoever they can find until each quota is filled.
Advantages: fast, cheap, and no population list is needed; ensures each group appears.
Disadvantages: not random, because the researcher chooses who to approach, so bias creeps in — approachable people are over-represented.

Cluster sampling

The population is divided into clusters (whole classes, whole suburbs, whole stores), a number of clusters are selected, and everyone inside the selected clusters is surveyed.
Advantages: very efficient when the population is spread out; no full list of individuals needed.
Disadvantages: people within a cluster tend to be similar, so the sample is less varied than its size suggests — three classes of Year 12 students are not a cross-section of a school.

Larger samples reduce the effect of unusual individual answers, but a large biased sample is worse than a small unbiased one — it is confidently wrong.
The standard asks for sufficient primary data of adequate quantity and quality. For a school-based project that usually means tens rather than hundreds, but it must be enough that one unusual respondent cannot move your conclusion.
Report the response rate, not just the number returned: 40 responses from 50 handed out is a much stronger result than 40 from 400.

Convenience bias — surveying whoever is nearby. Surveying at the canteen queue at lunchtime samples people who already buy lunch, which will overstate demand for a lunch product.
Self-selection bias — only people who care enough to respond do so, and they are usually the most enthusiastic or the most annoyed.
Time and place bias — a Friday afternoon sample is different from a Monday morning one.