What Are Confounding Variables?
A confounding variable is a factor related to both the exposure (or treatment) and the outcome that can distort the apparent relationship between them. It is not a step on the causal pathway; it offers an alternative explanation for an association.
A simple example
Illustrative example (invented). Observational data show that people who carry lighters have more lung cancer. Smoking is related to carrying lighters and also causes lung cancer, so smoking confounds the association. Controlling for smoking would remove the apparent link.
Conditions for confounding
A variable is a potential confounder if it (1) is associated with the exposure, (2) is associated with the outcome independently of the exposure, and (3) is not an intermediate step between exposure and outcome.
Ways to control confounding
| Stage | Method |
|---|---|
| Design | Randomisation; restriction (for example only non-smokers); matching |
| Analysis | Stratification; regression adjustment; propensity-score methods |
| Planning | Identify confounders in advance using subject knowledge, and measure them reliably |
Randomisation balances known and unknown confounders on average, which is why randomised trials are strong for causal inference. In observational studies, unmeasured confounders can remain; this is called residual confounding.
Confounders versus mediators and colliders
Adjusting for a mediator (on the causal pathway) can hide a real effect, and adjusting for a collider can create bias. Use a causal diagram to choose adjustment variables thoughtfully, with statistical guidance.
Practical steps for a student project
Before collecting data, list likely confounders from clinical knowledge and the literature (age, sex, comorbidity, severity, socio-economic status) and make sure they can be measured. Draw a simple diagram of how exposure, outcome and other variables relate. Record the confounders in your protocol and data dictionary, and ask a statistician how to adjust for them. State in the limitations which confounders you could not measure.
Common pitfalls
- Adjusting for everything available.
- Claiming that adjustment "eliminated" confounding.
- Ignoring confounding by indication in treatment studies.
Key takeaways
- A confounder is linked to both exposure and outcome and is not on the causal pathway.
- Control by design and analysis; acknowledge residual confounding.
References and further reading
- Hulley SB, Cummings SR, Browner WS, Grady D, Newman TB. Designing Clinical Research. 4th ed. Lippincott Williams & Wilkins; 2013. Link
- von Elm E, Altman DG, Egger M, et al. The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement: guidelines for reporting observational studies. BMJ. 2007;335:806-808. doi:10.1136/bmj.39335.541782.AD
- Sterne JA, Hernan MA, Reeves BC, et al. ROBINS-I: a tool for assessing risk of bias in non-randomised studies of interventions. BMJ. 2016;355:i4919. doi:10.1136/bmj.i4919
Related reading
This page is general educational information about research methods. It is not medical, legal or regulatory advice; follow your institution's ethics board and your target journal's instructions.