When Pearson Correlation Fails
What you'll learn: You'll discover the limitations of Pearson correlation and recognize situations where it gives misleading results.
The Problem with Pearson's r
Pearson correlation measures linear relationships between two continuous variables. But not all relationships in data follow a straight line, and not all data types are appropriate for this test. When these conditions aren't met, Pearson's r can dramatically underestimate—or completely miss—strong relationships.
Three Critical Failure Scenarios
Non-Linear Relationships
Imagine height and weight follow a curved pattern, or investment returns follow an exponential growth curve. Pearson's r might report a weak correlation (say, r = 0.3) even when the variables are perfectly related—just not in a straight line. Always check your scatterplot first.
Categorical Data
Pearson correlation requires continuous numerical data. If you try correlating categories like "small/medium/large" coded as 1/2/3, the result is meaningless. The numbers are arbitrary labels, not true measurements, so calculating r treats the equal spacing between categories as meaningful when it isn't.
Outliers and Violations
Remember from lesson 481 that outliers can drastically distort r. Similarly, if your data violates assumptions (like having a restricted range or severe skewness), Pearson's r may not accurately reflect the true relationship.
What to Do Instead
When Pearson fails, you need alternative methods: rank-based correlations (like Spearman's) for monotonic non-linear relationships, Chi-squared tests for categorical associations, or transformation of variables to linearize relationships. The key is recognizing when you're in one of these scenarios.
Key Takeaway: Pearson correlation only works for linear relationships between continuous variables; non-linear patterns, categorical data, and assumption violations require different correlation methods to avoid misleading conclusions.