A confounder is a third variable that is related to both an exposure and an outcome, making the exposure look more or less important than it really is.
A simple example
If supplement users report better sleep, the supplement may not be the cause. They may also exercise more, drink less alcohol, have higher income, track sleep more carefully, or seek health information more often. Any of those variables can travel with supplement use and sleep quality.
That is confounding: the exposure and outcome move together partly because another factor connects them.
Why observational studies are vulnerable
An observational study measures what people do without assigning the exposure. Researchers can adjust for measured variables, but they cannot adjust for variables they did not measure well or did not measure at all.
This is why a clean association is not the same thing as causal proof. Correlation metadata should state what was adjusted for and where residual uncertainty remains.
How randomization helps
A randomized controlled trial reduces confounding by assigning the exposure by chance. With enough participants, known and unknown causes tend to distribute more evenly between groups.
Randomization is not magic. Small samples, dropout, poor adherence, and unblinded assessment can still distort results.
How this appears in evidence grading
Confounding lowers confidence when a claim feeds recommendation ranking. The effect is strongest when the claim comes from lifestyle-adjacent data where many behaviors move together.
High-quality evidence quality metadata should identify likely confounders instead of hiding them behind a single confidence label.
Safety note
Confounding can make benefits look larger or harms look smaller. Treat causal claims from uncontrolled data with caution.