Reverse Causality

mechanism

An observed relationship can be genuine while its causal arrow points backward: what appears to be X producing Y may instead be Y producing X.

Among people who die from lung cancer, recent quitters can be unusually common. Read that correlation forward and quitting starts to look lethal. Read the timeline, and the explanation reverses: a cancer diagnosis prompted many of them to quit.

Turn the arrow around

Correlation tells you that two variables travel together; it does not tell you which one moved first. Reverse causality appears when the outcome changes the supposed cause—often through a diagnosis, warning, or other response. The resulting data preserve the relationship but conceal its sequence. Before accepting X → Y, you must therefore test Y → X and identify the causal mechanism that could carry each direction.

Where it shows up

Why recent quitters appear at risk

The lung-cancer case shows how selection by timing distorts interpretation. People who quit after diagnosis enter the category “recent quitters,” so that category contains people already made ill by cancer. The illness predicts the quitting—not the other way around.

Reversal is a hypothesis, not a reflex

Discovering that X → Y is unproven does not establish Y → X. The relationship might run both ways, as in bidirectional causality, or arise from a confounding variable. Reverse the arrow to test the story, not merely to replace one unsupported certainty with another.

Reconstruct the timeline

Tomorrow, when someone presents a correlation as an explanation, write both arrows—X → Y and Y → X—and ask what event would make Y change X. Then check whether the measurements distinguish people exposed before the outcome from people who changed only after it.

Episodes that teach this

The Great LIE - Correlation is Not Causation - FutureIQ 7,309 views
"lung cancer causes quitting" after diagnosis, so recent quitters are overrepresented among lung cancer deaths.

Examples & generalisations

Concrete examples: Correlation is not causation - do windmills cause wind