Confounding: when a third factor explains the link

A confounder is a third factor connected to both the thing being studied and the outcome being measured, so it can produce a statistical link between the two even where the first thing causes nothing at all. It is one of the most common reasons an observational study reports a connection that a randomized trial later fails to confirm.

A textbook illustration is coffee and heart disease. Early studies linked heavy coffee drinking to higher rates of illness, until researchers noticed that heavy coffee drinkers in those populations were also disproportionately heavy smokers. Smoking, not coffee, was driving much of the risk, and the coffee habit was simply riding along with it. Once studies adjusted for smoking, most of the coffee link faded.

Two traps follow. First, a confounder can create an association that is not really there, as with coffee and smoking, or mask one that is, if it pushes against the true effect. Second, researchers can only adjust for a confounder they thought to measure, so an unmeasured factor, sometimes called a lurking variable, can keep distorting a result no matter how careful the statistics look.

When a headline reports that one behavior is linked to a health outcome, ask what else tends to travel with that behavior, such as income, age, smoking, or existing illness, and whether the study adjusted for it. A randomized trial sidesteps the problem by assigning people to groups regardless of their other traits. Pair any observational link with a look at how it holds up once the obvious confounders are accounted for.