Posters · Glossary
What is a confounding variable?
A confounding variable is a third factor that is associated with both the exposure being studied and the outcome, and is not on the causal path between them. Because it travels with the exposure, it can create, hide or exaggerate an apparent effect. Smoking confounds an observed link between coffee drinking and lung cancer.
Most surprising findings in observational research turn out to be confounding. Knowing how to spot it, and how a study controlled for it, is the quickest way to judge a results panel.
Nuwan Madhusanka · Co-founder
4 min read · Published
| Type of variable | Relationship | Example | What to do in analysis |
|---|---|---|---|
| Confounder | Causes or is linked to both exposure and outcome, outside the causal path | Smoking in a study of coffee and lung cancer | Adjust for it, stratify, match or restrict |
| Mediator | Lies on the path from exposure to outcome | Weight gain between shift work and diabetes | Do not adjust if you want the total effect |
| Collider | Is caused by both exposure and outcome | Hospital admission in a study of two diseases | Do not adjust or select on it; doing so creates bias |
How confounding distorts a result
Suppose a study finds coffee drinkers get more lung cancer. Smokers in the sample also drink more coffee, and smoking causes lung cancer, so coffee looks harmful even if it has no effect. Smoking is the confounder: linked to the exposure, a cause of the outcome, and not a step between them. Confounding can also run the other way, hiding a real effect when the confounder pushes the outcome in the opposite direction.
Controlling it in design
The cleanest defence is randomisation, which in a large trial balances known and unknown confounders between groups. Observational studies use restriction, such as studying only non smokers, and matching, such as pairing each case with a control of the same age and sex. Both work only for confounders the researchers measured or thought of, which is why observational results are described as associations.
Controlling it in analysis
After data collection, researchers stratify, reporting the effect separately within levels of the confounder, or use multivariable models such as logistic or Cox regression that estimate the exposure effect with the confounders held constant. A review in the Gastroenterology and Hepatology from Bed to Bench journal sets out these methods and their limits. Residual confounding remains when a confounder was measured crudely or not at all, and no model can adjust for what was never recorded.
Reading it on a poster
Look for a crude estimate beside an adjusted one, a list of what the model adjusted for, and a sentence on why. A large change between crude and adjusted estimates shows confounding was present and matters. A results panel that reports only one estimate, without saying what it adjusted for, leaves the reader unable to judge. Watch for adjustment for mediators, which removes part of the effect being studied, and for selection on a collider, which creates associations that are not real.
Showing it on a poster board
The poster tab has 40 A0 boards, every one with a figure block, section slots and an author block, and stat blocks on 26 of them, so a crude and an adjusted estimate can sit side by side as key numbers with the adjustment set in the line beneath. Only three boards carry a table, which suits listing the confounders and their prevalence in each group. The cohort study example reports a crude and an adjusted hazard ratio together, which is the pattern worth copying.
Questions people ask
Is age always a confounder?
Not always, but it often is, because age affects both many exposures and many health outcomes. Whether it confounds a particular study depends on whether it is linked to the exposure in that sample and causes the outcome. Most clinical and epidemiological studies adjust for age by default and should say so.
How do researchers decide which variables to adjust for?
Ideally from prior knowledge of the causal structure, often drawn as a causal diagram, rather than by testing which variables change the result. Variables chosen only because they were statistically significant can include mediators or colliders. A good methods section names each adjustment variable and explains why it was included.
Can a randomised trial have confounding?
Randomisation balances confounders on average, so large trials are largely protected. Small trials can be unbalanced by chance on an important factor, and problems after randomisation, such as people switching groups or dropping out unevenly, can reintroduce bias. Trials therefore still report baseline characteristics by group.
What is residual confounding?
The confounding that remains after adjustment, because a confounder was measured imprecisely, grouped too broadly, or not measured at all. Adjusting for smoking as yes or no, for example, leaves residual confounding by how much people smoke. Authors should discuss it as a limitation rather than treating the adjusted estimate as final.
What is the difference between confounding and effect modification?
Confounding distorts the overall estimate and should be removed. Effect modification means the effect genuinely differs between groups, such as a drug working better in younger patients, and should be reported, not adjusted away. Stratified results help tell them apart: similar effects within strata but a different crude estimate suggest confounding.
Can a confounder be a variable nobody thought to measure?
Yes, and that is the main limitation of observational studies. An unmeasured confounder cannot be adjusted for, so authors sometimes estimate how strong such a factor would need to be to explain away the result. Stating the important variables that were not measured is part of an honest limitations section.
Make one with posters
The button opens the generator with this use case already described. Change the wording to match your own.
Create a poster with OneCraftRelated questions
- Cohort vs case-control studyCohort vs case control study: a cohort follows people from exposure to outcome, a case control study starts at the outcome and looks back. What each estimates.
- What is a hazard ratio?A hazard ratio compares how fast an event happens in two groups over time. What HR 0.6 or 1.5 means, how to read its interval, and how to report it on a poster.
- What is an odds ratio?What is an odds ratio: the odds of an outcome in one group divided by the odds in another. A worked 2x2 example, how to read it, and why it is not a risk ratio.
- What is intention to treat?What is intention to treat: analysing trial participants in the groups they were randomised to, whatever they received. Why it matters, and ITT vs per protocol.
Step by step in the builder: Add charts and diagrams.
Written and checked by the OneCraft team. Last checked .