Limitations of Pearson Correlation
You'll learn why Pearson correlation can mislead you in certain data situations.
Pearson correlation is powerful, but it's not a universal tool. It measures only linear relationships and makes assumptions that don't always hold in real-world data. Understanding when it fails prevents serious analytical mistakes.
Non-Linear Relationships
Pearson's r only detects straight-line patterns. Imagine a perfect U-shaped curve—as x increases, y decreases then increases. Two variables could be perfectly related, yet Pearson's r might be near zero! The relationship exists, but it's curved, not straight. Think of it like trying to measure the distance of a winding river using only a straight ruler—you'll miss most of what's really there.
Outliers Strike Again
As you learned earlier, outliers dramatically distort Pearson's r. A single extreme point can make uncorrelated data appear correlated, or mask a genuine relationship. Because the formula squares deviations, outliers get amplified. Always visualize your data with a scatterplot before trusting the correlation coefficient.
Ordinal Data Problems
Pearson correlation assumes your data is interval or ratio—meaning the distances between values are meaningful. With ordinal data (rankings like "1st place, 2nd place, 3rd place"), the gaps aren't necessarily equal. Using Pearson's r here treats "the difference between 1st and 2nd" the same as "between 99th and 100th," which may not reflect reality. This is why Spearman's rank correlation exists (though you haven't learned it yet).
Restriction of Range
If you only examine a narrow slice of your data, correlation weakens artificially. Measuring height and weight only among NBA players would show less correlation than in the general population—everyone's already tall and heavy!
Key Takeaway: Pearson correlation only measures linear relationships in continuous data and is vulnerable to outliers, non-linearity, and improper data types—always visualize first and understand your data's structure.