A sample is representative only in relation to a population and a question. Matching customers by age and region may be useful, but it does not establish that the sample also reflects their product experience, access needs or likelihood of leaving. The word should therefore begin a discussion of recruitment rather than end it.
Researchers use “representative sample” to describe a sample that supports a reasonable account of the target population on relevant characteristics or outcomes. Representativeness is not guaranteed by size, a demographic quota or a panel provider’s label. The route into the sample and the assumptions behind the analysis remain consequential.
Name the population before choosing the sample
A fictional music service wants to understand listening during a free trial. Current paying customers are easy to contact, but they have already passed through a selection process that trial users have not. Their account of the trial may omit reasons other people left and may reflect a product version that has since changed.
Define the target group and period precisely, then identify the characteristics likely to matter. In this case, trial stage, device, usage and completion status may be more directly relevant than a broad national demographic profile. The recruitment frame should offer an appropriate route to those people.
The Australian Bureau of Statistics’ guidance on sample design distinguishes probability and non-probability methods. Those methods provide different grounds for drawing conclusions beyond the participants studied.
Check more than the achieved totals
Compare the sample with relevant information about the target population where available. Examine who was eligible, who could be reached and who responded. Matching visible characteristics is reassuring only to the extent that they address plausible sources of distortion; unmeasured differences may remain.
Random sampling can support population inference under the design, but coverage and non-response still need attention. Quotas and weighting can address selected imbalances under assumptions. They do not make every difference disappear or automatically turn a convenience sample into a probability sample.
A larger sample generally improves precision under an appropriate model, but it does not repair systematic exclusion. Recruiting more paying customers would not solve the music service’s need to understand people who left during the trial.
Ask whether population estimation is the purpose
A small qualitative study may deliberately select people with particular experiences to understand how a problem occurs. It should include participants relevant to the decision, but it need not imitate the population’s proportions to be useful. Purposive sampling has a different purpose from estimating how common an opinion is.
The distinction should survive reporting. Describe the sample and recruitment method, explain known gaps and match the claim to the evidence. “These participants could not find the offline setting” may justify a design investigation; “most trial users cannot find it” requires a stronger basis for estimating prevalence.
A useful statement of representativeness tells the reader who the findings concern, what has been checked and where uncertainty remains. Without those details, the label offers more confidence than information.