Calculate Pearson's r with step-by-step computation, Fisher z confidence intervals, t-test significance, covariance, and full deviation table.
Pearson's r measures the strength and direction of a linear relationship between two variables. This calculator shows the full computation path instead of only the final coefficient, so you can inspect the deviations, cross-products, and sums of squares that build the result.
Alongside r, the page also reports R², a t-test for significance, covariance, and Fisher z confidence intervals. That makes it useful when you need both the coefficient and the uncertainty around it.
Preset datasets cover classic positive and negative relationships so the calculation steps are easy to compare against a known pattern.
Pearson correlation is often the first pass when you want to know whether a relationship is strong enough to matter and straight enough to model. Showing the arithmetic step by step makes it easier to audit the answer and explain where it came from.
The Fisher z interval and significance test add the uncertainty that a single r value cannot show on its own.
r = Σ(xᵢ−x̄)(yᵢ−ȳ) / √[Σ(xᵢ−x̄)²·Σ(yᵢ−ȳ)²]. t = r√(n−2)/√(1−r²), df=n−2. Fisher z = ½·ln((1+r)/(1−r)), SE_z = 1/√(n−3).
Result: r = 0.9972, R² = 0.9945, t = 35.58 (p < 0.001), 95% CI [0.9872, 0.9994]
Height and weight show a very strong positive linear correlation. 99.45% of weight variation is linearly associated with height. The Fisher z CI confirms the true population r is between 0.987 and 0.999.
Pearson's r is the ratio of covariance to the product of standard deviations: r = Cov(X,Y)/(SD_X · SD_Y). Expanding Cov(X,Y) = Σ(xᵢ−x̄)(yᵢ−ȳ)/(n−1), we get the familiar formula. The denominator normalizes the covariance to the [−1, +1] range regardless of variable scales.
If all points fall exactly on a line with positive slope, every (xᵢ−x̄)(yᵢ−ȳ) term is positive, and r = +1. If the line has negative slope, they're all negative, giving r = −1. Scattered points produce a mix of positive and negative terms that partially cancel, yielding |r| < 1.
The null hypothesis H₀: ρ = 0 is tested using t = r√(n−2)/√(1−r²) with n−2 degrees of freedom. Reject H₀ when |t| > t_critical. For testing H₀: ρ = ρ₀ (some non-zero value), convert to z-scores: z = (z_r − z_ρ₀) / SE_z, where z_r = Fisher transform of r and SE_z = 1/√(n−3).
In psychology and social sciences, Cohen's guidelines classify r = 0.10 as small, 0.30 as medium, 0.50 as large. In medical research, r = 0.30 might be clinically meaningful. In physics, r < 0.99 might indicate measurement error. Always interpret r in context, not by universal cutoffs.
Last updated:
Pearson's r measures the strength and direction of linear association. It ranges from −1 (perfect negative) through 0 (no linear relationship) to +1 (perfect positive). Only linear relationships are captured — a perfect parabola gives r ≈ 0.
Raw r has a skewed sampling distribution, especially near ±1. Fisher's z-transform converts r to a normally distributed z-score, enabling accurate confidence intervals and hypothesis tests for the population correlation ρ.
Ideally: (1) both variables are continuous, (2) relationship is linear, (3) no extreme outliers, (4) bivariate normality for significance tests. It's robust to mild violations of normality with n ≥ 30.
Pearson measures linear association using raw values. Spearman measures monotonic association using ranks. For linear relationships, both give similar results. For nonlinear monotonic relationships (log, exponential), Spearman is higher.
CI width depends on n and |r|. With n=10 and r=0.50, a 95% CI might be [−0.17, 0.87]. You need about n=40 for useful CIs with moderate correlations. Larger samples → narrower CIs.
Yes. A single extreme point can inflate r from 0.1 to 0.8 or deflate it from 0.9 to 0.3. Always check for outliers before interpreting Pearson's r, or use Spearman's rank correlation instead.