13 Research methods, statistics, critical review and evidence-based practice13.8 Descriptive statistics
13.8.1 Samples and populations
2 min read
Written by Resident Doctor Committee · Published
New in the 2027 syllabus
The College added this topic for exams from January 2027.
Who is being studied
Term | Meaning | Example |
|---|---|---|
Target population | The people the research question is about | All adults using a region's community mental health services |
Source population | The people you can reach through the recruitment setting | Adults registered with participating teams |
Sampling frame | The list people are selected from | A register of eligible patients |
Sample | The people selected for observation | Those chosen from that register |
Participants | The selected people who actually take part | Those who consent and provide data |
A parameter describes the population, such as its mean symptom score. A statistic describes the sample and can estimate the parameter.
A census tries to observe the whole defined population. Non-response and measurement error can still affect it.
Precision and representation
A sample can mislead through chance or through bias.
Sampling variability: a statistic differs by chance from one sample to the next, even when every sample uses the same sound method.
Larger samples usually give more precise estimates. Making a biased sample bigger does not remove its selection bias.
Coverage error happens when the sampling frame leaves out part of the target population, such as people outside registered services.
Non-response bias happens when taking part is linked to the thing being measured and the analysis does not properly allow for this.
Selection is not allocation
Random sampling selects people from the population with known probabilities of selection. This supports estimates about the population, provided coverage and response are adequate.
Random allocation assigns recruited participants to interventions. On average it reduces confounding of the treatment comparison.
A randomised trial can therefore have strong internal validity (a fair comparison of treatments) while its sample differs from routine patients.
Generalisability, whether the results apply to your population, depends on who was included, who was missed, and whether the factors that affect the result differ in the population you care about.