Sampling and Representativeness

#Distinguish the target population from the sample

The target population is the group a study aims to inform. The sample is the group actually included. These may differ because of location, eligibility rules, willingness to participate or access to care. Define both clearly so readers can judge whether the evidence fits the population of interest.

Recruitment methods matter. Random selection, consecutive recruitment and volunteer recruitment create different opportunities for selection bias. A sample drawn from one specialist clinic may differ from people seen in general care. Having many participants does not automatically correct a recruitment process that systematically leaves relevant people out.

#Describe who enters and who is missing

Report how people were identified, approached, assessed for eligibility and included. Where information is available, describe nonparticipation and exclusions. Follow-up losses also matter: people who remain in a study may differ from those who leave, particularly if leaving is related to health or the outcome being measured.

Assess characteristics relevant to the intended use, not just easily recorded categories. Disease severity, other conditions, language, equipment and care pathways can affect findings. The proportion of people with the condition also affects some test-performance measures. Privacy and ethical limits may constrain what can be collected or reported.

#Make cautious claims about wider relevance

Representativeness is not a single property that a sample either has or lacks. A group may be suitable for one question but poorly suited to another. Explain which features need to resemble the target population and which differences could plausibly change the result or its interpretation.

Statistical weighting or adjustment may reduce some differences when suitable information and assumptions are available. These methods cannot reliably recover missing perspectives or remove every unmeasured difference. Report remaining uncertainty and avoid extending conclusions to groups or settings that were not adequately studied without further supporting evidence.

#Common misunderstandings

A large sample is not automatically representative. Recruiting more people can improve precision, but it does not necessarily correct a recruitment process that leaves some groups out. A study with thousands of volunteers may still mainly describe people who had the time, resources or interest to take part.

Random sampling and random assignment are also different. Random sampling concerns who is selected from a population. Random assignment concerns which treatment or comparison group enrolled participants enter. Random assignment can strengthen a treatment comparison without making the participants representative of everyone who might receive that treatment.

Nor does a sample need to match every population characteristic for its findings to be useful. Differences matter especially when they could affect the outcome or treatment response. Finally, reporting similar results across subgroups does not prove that effects are identical: small subgroup numbers can leave important differences uncertain.

#Questions worth asking a clinician

• Who was this study intended to represent, and how does the enrolled sample differ from that population and from patients like me?
• How were participants recruited and selected, and could those methods have introduced selection bias despite a large sample?
• What do you know about people who declined participation, and how might their absence affect the findings?
• How many participants were lost to follow-up, and could differences between those who left and those who stayed affect the results?
• Which differences between me and the study participants matter most when deciding whether these findings apply to my care?