Pearson Correlation

The Pearson correlation coefficient (r) is a measure of the strength and direction of the linear relationship between two continuous variables.^[1] It ranges from negative 1 to positive 1, where negative 1 indicates a perfect negative linear relationship, positive 1 indicates a perfect positive linear relationship, and 0 indicates no linear relationship.

The Pearson correlation is one of the most frequently computed statistics in research, used in virtually every field that examines relationships between quantitative variables. It provides a standardized, easily interpretable measure of association that allows researchers to quantify how strongly two variables move together. From predicting academic performance to evaluating the relationship between biomarkers and disease progression, the Pearson correlation is often the first step in understanding bivariate relationships and serves as the foundation for more complex techniques such as regression analysis, factor analysis, and structural equation modeling. Its ubiquity makes it essential that researchers understand both its proper application and its limitations.

The Pearson correlation coefficient is calculated as the covariance of two variables divided by the product of their standard deviations.^[1] In simpler terms, it measures how much two variables vary together relative to how much they vary individually. If high values of one variable tend to co occur with high values of the other (and low with low), the correlation is positive. If high values of one variable tend to co occur with low values of the other, the correlation is negative.

Interpretation of the magnitude of r depends on the research context, but Cohen (1988) proposed conventions that classify correlations around 0.10 as small, 0.30 as medium (often called moderate), and 0.50 as large, intended for use when no better basis for judging the size of an effect is available.^[4] These benchmarks should be applied judiciously; in some fields, a correlation of 0.30 may be considered substantial if it predicts an important outcome, while in others, correlations below 0.70 may be considered insufficient for practical purposes such as measurement reliability.

The coefficient of determination, r squared, indicates the proportion of variance in one variable that is accounted for by the other.^[1] A correlation of r = 0.50 means that 25% of the variance in one variable is shared linearly with the other, while the other 75% is not accounted for by a linear relationship with it. This squared interpretation helps prevent overestimation of the strength of the relationship.

When these assumptions are violated, alternatives include the Spearman rank correlation (for ordinal data or nonlinear monotonic relationships), the Kendall tau correlation (for small samples with many tied values), and the point biserial correlation (for the relationship between a continuous and a dichotomous variable).

Use the Pearson correlation when you want to quantify the linear relationship between two continuous variables that meet the assumptions of linearity, normality, and homoscedasticity. Always create a scatterplot before computing the correlation to visually inspect the relationship for linearity, outliers, and restriction of range.^[1][3] Report r with its 95% confidence interval and sample size, and interpret the effect size in the context of the research question rather than relying solely on statistical significance.^[2]

A health psychology researcher examines the relationship between daily step count (measured by an accelerometer) and self reported stress (measured by the Perceived Stress Scale) in 300 university employees. A scatterplot shows a roughly linear relationship with no extreme outliers. The Pearson correlation is r = negative 0.35 (95% CI: negative 0.45 to negative 0.25, p < 0.001), indicating a moderate negative linear relationship: higher daily step counts are associated with lower perceived stress. The r squared of 0.12 indicates that 12% of the variability in perceived stress is accounted for by step count. The researcher appropriately notes that the cross sectional design precludes causal inference and that the relationship may be confounded by variables such as job type and physical health status.