Selection bias
- Bias
It happens when the people who get into or stay in a study do not represent the population of interest, and that distorts the result.
What it is
Selection bias arises when the way participants are recruited or retained makes the relationship between exposure and outcome different in the study from that in the population of interest. It includes Berkson bias (using hospital patients), the healthy worker effect (comparing workers with the general population), volunteer bias and non-response. It is hard to fix in the analysis: it is prevented at the design stage.
How it arises
- The chance of entering the study depends on both exposure and outcome.
- In case-control studies: poorly chosen controls (for example, from a hospital where the exposure is more common).
- In cohorts: losses to follow-up that differ between groups.
- In surveys: non-responders differ from responders.
Worked example
- 4 000 factory workers are followed for 10 years: 120 die.
- National rates for age and sex predicted 160 deaths.
- Standardised mortality ratio (SMR) = observed / expected = 120 / 160 = 0.75.
- Working at the factory seems protective: 25% fewer deaths.
- But to be hired and keep working one must be healthy; the general population includes the sick and disabled.
- Against workers in another industry, 115 deaths were expected: SMR = 120 / 115 ≈ 1.04.
- Reading: the apparent protection was the healthy worker effect, a selection bias.
Common forms of selection bias
| Type | What happens | Where it appears |
|---|---|---|
| Berkson (hospital admission) | Admission depends on both exposure and disease | Case-control studies with hospital controls |
| Healthy worker | Workers are healthier than the general population | Occupational cohorts |
| Volunteer | Volunteers have other habits and better health | Surveys, trials, screening |
| Non-response | Non-responders differ from responders | Surveys |
| Loss to follow-up | Drop-outs differ by group | Cohorts and trials |
Markers
- SMR
- Berkson
How to prevent or control it
- Choose controls from the same population that produced the cases, with the same chance of being included.
- In occupational cohorts, compare with another working population, not the general population.
- Aim for high response rates and compare responders with non-responders.
Common pitfalls
- Confusing selection bias with confounding: confounding can be adjusted in the analysis; biased selection almost never.
- Thinking a large sample makes up for a badly chosen one.
Test yourself
Think of the answer, then tap the question to see it.
What is the healthy worker effect?
The lower mortality of workers compared with the general population, because one must be fairly healthy to work.
What is Berkson bias?
A selection bias in studies of hospital patients: the chance of admission depends on both the exposure and the disease, distorting the association.
Can selection bias be corrected by adjusting in the analysis?
Usually not; it must be prevented at design by choosing and retaining participants well.
Connections
Educational content on epidemiological and statistical methods; the examples use made-up numbers to teach the calculation. Not a clinical or legal guide: current Colombian rules are in the official sources.
Sources: Selection bias | Catalog of Bias (Catalogue of Bias Collaboration, 2017); Admission rate bias | Catalog of Bias (Catalogue of Bias Collaboration, 2018); Chapter 25: Assessing risk of bias in a non-randomized study (Cochrane, 2024).
Spotted a mistake?We are always improving for the community. Tell us what is wrong and our team will review it.Report a mistake
Representative model, for educational purposes only.