🧪 This platform is in early beta. Features may change and you might encounter bugs. We appreciate your patience!
correlation_causation_fallacy
The correlation-causation fallacy is the error of inferring a causal relationship between two variables solely because they are statistically correlated. A correlation can arise from direct causation, reverse causation, confounding, or chance. It is distinct from the reverse causality fallacy and from spurious correlation, though all three involve misinterpreting correlational evidence.
Cities with more hospitals have higher death rates. Does building hospitals cause death? No — more hospitals are built in cities with more sick people. The causal arrow runs from disease burden to hospital construction, while death follows from disease, not from hospitals.
Data show that children who have more books in their home score higher on literacy tests. A school district launches a program to distribute free books to low-income households, expecting test scores to rise automatically. The correlation likely reflects parental education and engagement — simply adding books without addressing those underlying factors produces little effect.
A study finds that people who carry lighters are significantly more likely to develop lung cancer. A naive reading suggests lighters cause cancer. In reality, carrying a lighter is a proxy for smoking behavior — the lighter is correlated with cancer only because it is associated with the true causal agent.
Binary (yes/no) questions an LLM must answer to identify this aspect:
Does the argument infer a causal relationship from an observed correlation without additional evidence?
Type: binaryHave plausible confounders been controlled for in the analysis?
Type: binaryIs reverse causality (the effect causing the exposure) being ruled out?
Type: binaryIs a plausible biological, social, or mechanical mechanism identified and tested?
Type: binaryThe correlation-causation fallacy is the error of inferring a causal relationship between two variables solely because they are statistically correlated. A correlation can arise from direct causation, reverse causation, confounding, or chance. It is distinct from the reverse causality fallacy and from spurious correlation, though all three involve misinterpreting correlational evidence.
Causal narratives are easier to comprehend and remember than correlation-only descriptions. The human mind is strongly disposed to construct causal stories from co-occurrence patterns.
Require a plausible causal mechanism. Check whether the association persists after controlling for confounders. Evaluate temporal order. Look for natural experiments, instrumental variables, or randomized evidence.
Ice cream sales and drowning deaths are correlated (both peak in summer). Correlation-causation reasoning would absurdly imply ice cream causes drowning. The actual cause is hot weather and increased outdoor activity.
Use these tools to detect, analyze, or train this aspect.