Methods of sampling
Studying this course? Keep your progress in one place.
Key facts
- A census is exact but slow, costly and impossible if testing destroys the item; a sample is cheaper but has sampling error and may be biased.
- Simple random samples need a sampling frame and random numbers.
- Systematic: every th unit with and a random start.
- Stratified: random samples from each stratum, in proportion to its size.
- Quota: the interviewer fills quotas; it needs no frame but is non-random.
Census or sample
A census is exact but slow and costly; a sample is cheaper but has sampling error.
A population is the whole set under investigation; a census measures every member; a sample only some. A sampling frame is a list of all the sampling units, and a good frame is complete, up to date and free of duplicates.
A census has no sampling error but is impossible if testing destroys the item. A sample is cheaper, but there is a risk of bias if it is not representative.
- 1
Population
the whole set under investigation
- 2
Sampling frame
a list of all the sampling units
- 3
Sample
only some members are measured
- 4
Conclusions
about the population, with sampling error and possible bias
| Census | Sample | |
|---|---|---|
| Accuracy | True value, no sampling error | Sampling error, possible bias |
| Cost and time | Expensive and slow | Cheaper and quicker |
| Destructive testing | Impossible | The only option |
Census
- Accuracy:
- True value, no sampling error
- Cost and time:
- Expensive and slow
- Destructive testing:
- Impossible
Sample
- Accuracy:
- Sampling error, possible bias
- Cost and time:
- Cheaper and quicker
- Destructive testing:
- The only option
A company wants to know how long its batteries last. Why should it use a sample rather than a census?
Simple random sampling
Every possible sample of size is equally likely; it needs a sampling frame.
In a simple random sample of size , every possible sample of units is equally likely, so every unit has the same chance. Number the units, generate random numbers, and discard unusable numbers and repeats (selection is without replacement).
It is free of selection bias, but needs a complete frame and can be slow for a large population. A small sample may by chance miss a small group.
- 1
Number the units
Using the frame
- 2
Generate random numbers
From a calculator or table
- 3
Discard unusable numbers
Out of range, or repeats
- 4
Continue
Until units are chosen
A frame has 500 units numbered 001 to 500. A random number of 731 is generated. What do you do?
Systematic sampling
Take every th unit from the frame after a random start.
A systematic sample takes every th unit, where the sampling interval is . The random start is chosen from the first units, then every th unit after it is selected.
It is quick and spreads the sample evenly, needing only a random start. But it is not a simple random sample, since neighbouring units can never both be chosen, and a regular pattern in the list matching can cause bias.
Worked example
A list has students and a sample of is needed. The random start is . Which students are selected?
- 1
.
- 2
Start at the th name.
- 3
Add each time:
A frame has units and a sample of is needed. What is the sampling interval?
Stratified sampling
Take a simple random sample from each group, in proportion to its size.
When the population divides into distinct groups (strata) such as year groups, a stratified sample takes a simple random sample from each stratum, in proportion to its size. Every stratum is then represented properly, so it is more representative than a simple random sample.
It needs a complete list of each stratum, which costs time, and rounding may make the proportions slightly inexact.
- In the population
- In the sample (n = 50)
- Number from a stratumstratum size ÷ population × n
Worked example
A firm has employees in three departments, with , and staff. A stratified sample of is needed. How many come from each department?
- 1
Sampling fraction .
- 2
, , .
- 3
Check: .
A school has pupils in Year 10 and in Year 11. A stratified sample of is taken. How many are from Year 11?
Quota sampling
An interviewer fills fixed quotas in each group, so it is non-random and needs no frame.
In a quota sample the population is divided into groups such as age and gender, and an interviewer chooses people until each quota, proportional to the group size, is filled. It is non-random: the interviewer decides whom to approach.
It needs no frame, is quick and cheap, and refusals are simply replaced. But there is interviewer bias and selection bias, and sampling error cannot be estimated.
- 1
Divide into groups
for example by age and gender
- 2
Set quotas
proportional to the size of each group
- 3
Interviewer chooses people
non-random, no sampling frame needed
- 4
Stop when quotas are full
refusals are simply replaced
| Stratified | Quota | |
|---|---|---|
| Selection within groups | Random, from a sampling frame | By the interviewer, not random |
| Sampling frame | Needed | Not needed |
| Sampling error | Can be estimated | Cannot be estimated |
Stratified
- Selection within groups:
- Random, from a sampling frame
- Sampling frame:
- Needed
- Sampling error:
- Can be estimated
Quota
- Selection within groups:
- By the interviewer, not random
- Sampling frame:
- Not needed
- Sampling error:
- Cannot be estimated
Why can sampling error not be estimated for a quota sample?
Choosing a method
Match the method to the context, and link every advantage or disadvantage to it.
In an evaluation question give a point for the method and a point for the context, state a limitation, then reach a justified conclusion. A strong limitation is specific to the scenario, for example "only shoppers on a Saturday morning could be interviewed".
The method must give a representative sample.
- 1
Name the method
and link it to the context
- 2
Give an advantage
for example no frame needed, or every stratum represented
- 3
State a limitation
specific to the scenario
- 4
Justified conclusion
does the method give a representative sample?
| Best method | Because | |
|---|---|---|
| Complete list, small uniform population | Simple random | Needs frame and random numbers |
| Long ordered list, speed matters | Systematic (check for patterns) | Only a random start is needed |
| Distinct groups that may differ | Stratified | Each group is represented in proportion |
| No list available, quick and cheap | Quota | No sampling frame needed |
| Destructive testing or infinite population | A sample, never a census | Testing destroys the item |
Best method
- Complete list, small uniform population:
- Simple random
- Long ordered list, speed matters:
- Systematic (check for patterns)
- Distinct groups that may differ:
- Stratified
- No list available, quick and cheap:
- Quota
- Destructive testing or infinite population:
- A sample, never a census
Because
- Complete list, small uniform population:
- Needs frame and random numbers
- Long ordered list, speed matters:
- Only a random start is needed
- Distinct groups that may differ:
- Each group is represented in proportion
- No list available, quick and cheap:
- No sampling frame needed
- Destructive testing or infinite population:
- Testing destroys the item
A researcher wants opinions from shoppers in a street, and has no list of the population. Which method is most suitable?
Try an exam question
A sixth form has students in Year 12 and in Year 13. A stratified sample of students is taken. (a) Find the number of students from each year. (b) Give one advantage of stratified sampling over simple random sampling.
[4 marks]
- [1]Sampling fraction .
- [1]Year 12: students.
- [1]Year 13: students.
- [1]Each year is represented in proportion, so the sample is more representative.
That's the notes covered.
Carry on to the next subtopic.
Try a question
Which of the following is the sampling frame for this survey?
1 mark, marked against the mark scheme
Which of the following is the sampling frame for this survey?
1 mark, marked against the mark scheme