What if my dataset contains severe outliers?

Pearson's $r$ is highly sensitive to outliers. If removing the outlier is not mathematically justified, use a non-parametric alternative like Spearman's rank correlation coefficient.

Can I use this method for categorical data?

No. Pearson's correlation requires continuous, quantitative variables. For categorical data, use a Chi-square test of independence or point-biserial correlation, depending on the variable types.

What does an $r$ value of $0$ mean?

An $r=0$ strictly means there is no *linear* relationship between the variables. A strong non-linear relationship (like a parabola) might still exist.

How to calculate a correlation coefficient

Pearson's correlation coefficient ( $r$ ) quantifies the strength and direction of the linear relationship between two continuous variables. Use this method when you have paired quantitative data $(x_i, y_i)$ and need to measure linear association, but not causation or non-linear trends.

The setup

Identify your paired dataset consisting of $n$ pairs of observations, $(x_1, y_1), (x_2, y_2), \dots, (x_n, y_n)$ . Ensure both variables are continuous and quantitatively measured. You will need the sample means for both variables, denoted as $\bar{x}$ and $\bar{y}$ .

The steps

Calculate the sample means $\bar{x} = \frac{\sum x_i}{n}$ and $\bar{y} = \frac{\sum y_i}{n}$ .
Compute the deviations from the mean for every data point: $(x_i - \bar{x})$ and $(y_i - \bar{y})$ .
Calculate the sum of products of the deviations: $SP_{xy} = \sum (x_i - \bar{x})(y_i - \bar{y})$ .
Calculate the sum of squares for both variables: $SS_x = \sum (x_i - \bar{x})^2$ and $SS_y = \sum (y_i - \bar{y})^2$ .
Compute the correlation coefficient using the formula: $r = \frac{SP_{xy}}{\sqrt{SS_x \cdot SS_y}}$

Checking the result

Verify that your result falls within the bounds $-1 \le r \le 1$ . An $r$ value near $1$ indicates a strong positive linear correlation, a value near $-1$ indicates a strong negative linear correlation, and a value near $0$ indicates a weak or non-existent linear correlation.

Common errors

A frequent error is breaking the pairing of $(x_i, y_i)$ data during sorting or calculation. Another is interpreting a high correlation coefficient as proof of causation. Finally, computing Pearson's $r$ for data that has a clear non-linear relationship (e.g., quadratic) will yield a misleading near-zero value.

Worked example

Calculate the correlation coefficient for the paired data: $X = (1, 2, 3)$ and $Y = (1, 4, 4)$ .

Calculate $n$ : $n = 3$ .
Calculate means: $\bar{x} = \frac{1 + 2 + 3}{3} = 2$ . $\bar{y} = \frac{1 + 4 + 4}{3} = 3$ .
Calculate deviations $(x_i - \bar{x})$ : $-1, 0, 1$ .
Calculate deviations $(y_i - \bar{y})$ : $-2, 1, 1$ .
Calculate $SP_{xy}$ : $(-1)(-2) + (0)(1) + (1)(1) = 2 + 0 + 1 = 3$ .
Calculate $SS_x$ : $(-1)^2 + 0^2 + 1^2 = 1 + 0 + 1 = 2$ .
Calculate $SS_y$ : $(-2)^2 + 1^2 + 1^2 = 4 + 1 + 1 = 6$ .
Compute $r$ : $r = \frac{3}{\sqrt{2 \cdot 6}} = \frac{3}{\sqrt{12}} = \frac{3}{2\sqrt{3}} = \frac{\sqrt{3}}{2} \approx 0.866$ The correlation coefficient is approximately $0.866$ .

FAQ

Run your own problem

Answer form

Drag an image or PDF of the problem here, or click to upload

Ctrl + V to paste

References: OpenStax Introductory Statistics, Chapter 12: Linear Regression and Correlation · Moore, McCabe, and Craig: Introduction to the Practice of Statistics