Information (measurement) bias
- Bias
If exposure or disease is measured badly, people land in the wrong box and the association is distorted.
What it is
Information (measurement) bias occurs when exposure or outcome is recorded with error and participants are misclassified. If the error is the same in all groups (non-differential), it usually pulls the association towards the null (RR or OR towards 1). If the error differs between groups (differential), it can exaggerate or hide the association. It is prevented with validated instruments, clear definitions, objective records and blinded assessors.
How it arises
- The measuring tool has imperfect sensitivity or specificity.
- Some exposed people are classed as unexposed (or the reverse), or some sick people as healthy.
- If this happens equally in cases and controls, the error is non-differential.
- If it happens more in one group (for example, cases remember better), the error is differential.
Worked example
- True data from a case-control study: 50 of 100 cases and 25 of 100 controls were exposed.
- True OR = (50 × 75) / (50 × 25) = 3 750 / 1 250 = 3.0.
- Non-differential: the questionnaire picks up only 80% of the exposed in both groups: 40 cases and 20 controls exposed.
- Observed OR = (40 × 80) / (60 × 20) = 3 200 / 1 200 ≈ 2.7: closer to 1.
- Differential: cases recall 100% (50 exposed) and controls 80% (20 exposed).
- Observed OR = (50 × 80) / (50 × 20) = 4 000 / 1 000 = 4.0: further from 1.
- Reading: the same measuring error underestimates the OR if it is even and exaggerates it if it hits cases more.
Misclassification: types and effect
| Type | What happens | Usual effect on the RR or OR |
|---|---|---|
| Non-differential | The error is the same in all groups | Towards 1 (underestimates the association) |
| Differential | The error depends on the group | Towards or away from 1, unpredictable |
How to prevent or control it
- Use clinical, pharmacy or occupational records rather than memory alone whenever possible.
- Measure every group the same way: same questionnaire, same trained interviewer, same laboratory.
- Blind whoever measures the outcome to the exposure or group.
Common pitfalls
- Assuming non-differential misclassification always biases towards 1: with more than two categories or correlated errors it is not guaranteed.
- Calling a measurement error confounding: they are different problems handled differently.
Test yourself
Think of the answer, then tap the question to see it.
What effect does non-differential misclassification of a two-level exposure usually have?
It pulls the measure of association towards 1, that is, it underestimates the association.
What is differential misclassification?
A measurement error whose size depends on the group (cases or controls, exposed or not), which can exaggerate or hide the association.
What is a typical example of differential information bias in a case-control study?
Recall bias: cases remember their past exposures better than controls.
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: Misclassification 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.