Lecture Notes: Spurious Correlations

Introduction to Spurious Correlations

• Correlation is not causation
  • The idea that just because two variables show a correlation does not mean one causes the other.

Examples of Spurious Correlations

• Elijah Wood Movies and Orderlies in Oklahoma
  • Charts show correlation between these two unrelated variables.
  • GenAI's humorous explanation suggests Middle Earth's influence on healthcare.
• Air Pollution in San Diego and the Popularity of 'Kirk'
  • Close correlation tracked over time.
  • AI's explanation humorously connects pollution with naming trends.
• Petroleum Consumption in Azerbaijan and Farm Equipment Mechanics in Alabama
  • Demonstrates how unrelated variables show correlation.
  • Explanation involves a fictional economic chain reaction.
• Master's Degrees in Education and Google Searches for 'Gangnam Style'
  • Decrease in degrees correlates with less interest in the dance.
  • Fictional explanation ties educators' dance skills to trend.

Detailed Explanations

Why Spurious Correlations Occur

1. Data Dredging
   • Large datasets result in random, strong correlations.
   • Example: Comparing 25,237 variables can yield random matches.
2. Lack of Causal Connection
   • No direct link between variables, though might seem so.
   • Example: Using 'years' as a variable creates unrelated connections.
3. Observations Not Independent
   • Trend lines can form as sequential years aren't independent.
   • P-values might not reflect reality.
4. Y-Axes Doesn’t Start at Zero
   • Graphs can visually deceive.
   • Line graphs emphasize connections more than warranted.
5. Confounding Variables
   • Third variables might influence both variables.
6. Outliers
   • Anomalous data points can skew correlation strength.
7. Low n (Sample Size)
   • Few data points lead to misleading correlations.

Analysis Tools and Insights

• Links and methods provided to delve deeper into correlations.
• Python code available for personal calculation.

Closing Thoughts

• Spurious correlations highlight the importance of careful analysis and the potential pitfalls in interpreting data visually or statistically.