9.1.2, 1.2.4, 6.2.2, D.1.5 - Sampling and generalisation

Who should the sample represent?

Sampling begins by deciding whose behaviour the research question concerns.

The target population is the group to which the researcher wants to generalise. The sample is the smaller group actually studied. If a study concerns adults enrolled at one college, its target population is those adults, not automatically all adults in the country. Define the population by relevant age, location, enrolment or other criteria before choosing participants.

A sampling frame is the list or source from which participants can be selected. An up-to-date register of all eligible students could provide a frame; a list of library members would omit non-members. Even perfectly random selection from the library list cannot represent everyone if library use is related to the behaviour being measured. This is coverage bias in the frame.

Generalisability concerns how far findings apply beyond the studied sample, setting and conditions. A representative sample resembles its target population on characteristics relevant to the question. A larger sample reduces some chance variation but does not automatically remove a systematic selection bias. Asking 1,000 highly motivated volunteers can still miss adults uninterested in the topic. Conversely, a small carefully selected sample may be suitable for an exploratory question, while leaving wider estimates uncertain.

Distinguish the intended sample from the achieved sample. Some selected people may decline or leave. Record response and attrition without coercing participation; voluntary consent takes priority over forcing an exact sample. Generalise to a clearly stated population and acknowledge relevant exclusions rather than claiming universal behaviour.

Selecting with an equal chance

Random sampling uses a chance procedure rather than a researcher's personal choice of convenient participants.

For a simple random sample, every eligible person in the sampling frame has an equal chance of selection. Suppose a college has 200 eligible adults and needs 20 invitations. Give every adult a unique number from 1 to 200, then use a random-number generator to select 20 different numbers without replacement. Invite the corresponding people privately, explain consent and record who accepts. State the frame, numbering and selection rule so someone can reproduce the method, even though a fresh random draw need not select the same people.

Random selection reduces deliberate selection bias and makes chance rather than preference determine invitations. It does not guarantee that a small sample exactly mirrors every population characteristic: a chance draw can still contain more of one group. It also needs a suitable frame. If many invitees decline and those who accept differ systematically, the achieved sample may be biased despite a random invitation procedure. Plan any reserve invitations through the same random method rather than filling gaps with whoever is nearby.

Random sampling and random allocation answer different questions. Sampling chooses who is invited from a population. Allocation assigns recruited participants to experimental conditions. Randomly putting a convenience sample into two conditions improves the fairness of that comparison but does not turn the recruited sample into a random population sample.

Representing relevant strata

A stratified sample deliberately represents relevant population subgroups before chance selection occurs within them.

For proportional stratified sampling, divide the frame into non-overlapping strata, such as year groups, and sample each in its population proportion. Choose strata relevant to the research question, not categories simply because they are easy to record. Each eligible person must belong to one stratum for that selection, and all eligible people must be covered.

In an illustrative college population of 240 adults, 120 are first-year students, 80 are second-year students and 40 are third-year students. A sample of 30 should include: