Identifying Potential Causal Relationships
What you'll learn: How to evaluate whether a correlation between variables might actually indicate a causal relationship using three key criteria.
Beyond Correlation
You've already learned that correlation doesn't imply causation and explored spurious correlations, third variables, and reverse causality. But how do you identify when a correlation might represent genuine causation? Three criteria help you evaluate potential causal relationships:
The Three Key Criteria
1. Temporal Order
Cause must precede effect. If A causes B, then A must happen before B. This seems obvious, but it's essential to verify timing in your data. If ice cream sales and drowning rates both peak in summer, neither causes the other—they share a temporal pattern driven by season.
2. Mechanism (Plausibility)
There must be a believable pathway. Can you explain how one variable could influence the other? A credible biological, physical, psychological, or social mechanism strengthens the causal case. For example, smoking causing lung cancer has a clear biological mechanism through carcinogen exposure.
3. Consistency
The relationship should appear across different contexts. If multiple studies, datasets, or populations show the same association, it's more likely causal. One-off correlations in a single dataset are weak evidence; replicated findings across time and place are stronger.
Applying the Criteria
Think of these as filters, not proof. A relationship passing all three criteria is potentially causal but still requires rigorous testing (randomized experiments, causal inference methods). Failing any criterion should make you skeptical.
Analogy: These criteria are like detective work—temporal order is the timeline, mechanism is the motive, and consistency is corroborating witnesses. No single piece proves guilt, but together they build a case.
Key Takeaway: To identify potential causal relationships, verify temporal order (cause before effect), establish a plausible mechanism (how it could work), and confirm consistency (replication across contexts).