Course contentsShow
← Data Science
Lesson 479 of 2,14513. Correlation AnalysisFree lesson

Assumptions of Pearson Correlation

Linearity, continuous variables, absence of outliers, and approximate normality requirements.

Assumptions of Pearson Correlation

What you'll learn: The conditions your data must meet for Pearson's r to give you accurate and meaningful results.

Why Assumptions Matter

Just like ANOVA and t-tests have assumptions, Pearson correlation only works properly when your data meets certain requirements. Violating these assumptions can lead to misleading correlation coefficients—you might see a weak correlation when a strong relationship exists, or vice versa.

The Four Key Assumptions

1. Linearity
Pearson's r measures linear relationships only. If your variables are related in a curved or U-shaped pattern, Pearson's r will underestimate or completely miss the relationship. Always visualize your data with a scatterplot first!

2. Continuous Variables
Both variables should be continuous (measured on an interval or ratio scale). Using ordinal or categorical data with Pearson's r is inappropriate—you'd need different correlation methods for those data types.

3. Absence of Outliers
Extreme values can dramatically distort Pearson's r because it's based on means and standard deviations. A single outlier can artificially inflate or deflate your correlation coefficient, making it unreliable.

4. Approximate Normality
While Pearson's r itself is relatively robust, if you plan to test significance or construct confidence intervals, both variables should be approximately normally distributed. Severe skewness can affect your p-values and inference.

Think of It Like This

Imagine measuring how much taller plants grow with more sunlight. Pearson's r assumes: the relationship is a straight line (linearity), you measured height in inches not categories (continuous), no mutant plants that grew 10 feet (no outliers), and your data looks bell-curved (normality).

Key Takeaway: Pearson correlation requires linearity, continuous variables, no extreme outliers, and approximate normality for valid inference—always check these assumptions before trusting your correlation coefficient.