Lecture on Correlation vs. Causation
Introduction
- Speaker: Peter van de Ven
- Topic: Dangers of misinterpreting correlations as causations.
Example 1: Ice Cream and Drownings
- Plotting a graph of ice cream sales vs. drownings shows an upward trend.
- Incorrect conclusion: Ice cream causes drownings.
- Actual cause: Nice weather causes both increased swimming and ice cream sales.
- Lesson: Correlation does not imply causation.
Logical Mistake
- Jumping to incorrect conclusions about causality when seeing a correlation is a common logical error.
Example 2: Marriage and Men's Longevity
- Statistics show married men live longer than single men.
- Misinterpretation: Marriage causes longer life for men.
- True causation: Healthy, wealthy, and well-educated men (who have higher life expectancy) are more likely to marry.
- Takeaway: High life expectancy leads to higher marriage rates, not the other way around.
Example 3: Children Sleeping with Lights On
- 1999 study linked night lights to myopia in children.
- Initial advice: Turn off lights to prevent short-sightedness.
- Correction: Short-sightedness is genetic; short-sighted parents are likely to leave lights on and have short-sighted children.
- Lesson: Simple correlation mistake; causation was misunderstood.
Example 4: Self-Esteem and Academic Success
- 1970s research linked high self-esteem with good grades.
- Misconception: High self-esteem leads to good grades.
- Actual direction: Good grades boost self-esteem.
- Problem: Children with high self-confidence but no achievements end up with low self-esteem.
Broader Implications
- Incorrect correlations have been drawn in various contexts (e.g., vaccines and autism, female bankers and financial crises).
- Strong correlation is insufficient to prove causation.
- Requirement for causation: Understand the why and how of the relationship.
Conclusion
- Always question causal claims based on correlation.
- When in doubt, remember the ice cream analogy.
Note: These notes summarize key points from a lecture on understanding the difference between correlation and causation, using various illustrative examples.