Paste two lists of paired values. The calculator returns the sample covariance (÷(n − 1)), the population covariance (÷n), Pearson’s r, both standard deviations and a scatter plot with the fitted line.
Worked steps
Enter up to four variables. The calculator returns the full symmetric sample covariance matrix, plus each variable’s mean and sample standard deviation.
Worked steps
The covariance calculator works out sample and population covariance for paired data, shows the correlation alongside, and plots the scatter. A covariance-matrix mode handles three or four variables at once. Covariance measures direction and strength in the units of your data, so the page explains why its size alone is not comparable.
What is covariance?
Covariance measures how two variables move together. For each pair, it multiplies how far x is from its mean by how far y is from its mean, then averages those products. When x is above its mean at the same time y is above its mean, the product is positive; when one is above and the other below, the product is negative.
The sign is the headline: a positive covariance means the variables tend to rise together, a negative covariance means one tends to fall as the other rises, and a covariance near zero means there is no linear pattern between them. Covariance does not prove that one variable causes the other — it only describes the linear association in the data you have.
Covariance is the building block of two familiar quantities. Divide it by the two standard deviations and you get Pearson’s correlation coefficient r; divide it by the variance of x and you get the slope of the least-squares regression line. That is why covariance matters even though you will rarely quote it on its own.
How to use the covariance calculator
Paired covariance. Paste your x values in the first box and your y values in the second, in the same order. The calculator returns the sample covariance, the population covariance, Pearson’s r, both standard deviations, a pair-by-pair deviation table and a scatter plot with the fitted line. Try it now with x = 12, 15, 18, 21, 24 and y = 30, 38, 44, 55, 61: the sample covariance is 59.2500 and r is 0.9959.
Covariance matrix. Choose 2, 3 or 4 variables, name each one and paste its values. The calculator returns the full symmetric sample covariance matrix, plus each variable’s mean and sample standard deviation. Try it now with study_hrs = 5, 6, 7, 8, 9, 4, sleep_hrs = 7, 6, 8, 7, 9, 6 and score = 80, 85, 88, 92, 95, 78: the matrix is [[3.5000, 1.7000, 12.4000], [1.7000, 1.3667, 5.7333], [12.4000, 5.7333, 44.2667]].
The formula — sample vs population
For paired data (x₁, y₁), …, (xₙ, yₙ) with means x̄ and ȳ, the sample covariance is:
Cov(x, y) = Σ(xᵢ − x̄)(yᵢ − ȳ) ÷ (n − 1)
The population covariance uses the same numerator but divides by n instead of n − 1:
Covpop(x, y) = Σ(xᵢ − x̄)(yᵢ − ȳ) ÷ n
Use the sample version when your data are a sample drawn from a larger population (the usual case, and the default on this page). Use the population version only when your data are the entire population you care about. The two differ by a factor of n/(n − 1), so they are close when n is large and noticeably different when n is small.
The regression link is just as direct: the slope of the least-squares line of y on x is slope = Cov(x, y) ÷ Var(x). If the covariance is positive the line slopes up; if it is negative the line slopes down. See the linear regression calculator to fit that line.
Worked examples
Example 1 — paired covariance
x = [12, 15, 18, 21, 24], y = [30, 38, 44, 55, 61]. The means are x̄ = 18.0000 and ȳ = 45.6000. The deviations multiply out to 93.6000, 22.8000, 0.0000, 28.2000 and 92.4000, which sum to 237.0000. The sample covariance is 237.0000 ÷ 4 = 59.2500; the population covariance is 237.0000 ÷ 5 = 47.4000. The sample standard deviations are SDx = 4.7434 and SDy = 12.5419, so Pearson’s r = 59.2500 ÷ (4.7434 × 12.5419) = 0.9959. The check: 0.9959 × 4.7434 × 12.5419 = 59.2500.
Example 2 — covariance matrix
Three variables measured on six students: study_hrs = [5, 6, 7, 8, 9, 4], sleep_hrs = [7, 6, 8, 7, 9, 6], score = [80, 85, 88, 92, 95, 78]. The sample covariance matrix is:
| study_hrs | sleep_hrs | score | |
|---|---|---|---|
| study_hrs | 3.5000 | 1.7000 | 12.4000 |
| sleep_hrs | 1.7000 | 1.3667 | 5.7333 |
| score | 12.4000 | 5.7333 | 44.2667 |
The diagonal holds the variances: study_hrs 3.5000, sleep_hrs 1.3667, score 44.2667. Off the diagonal, study time and score move together most strongly (12.4000), and sleep and score move together too (5.7333). All three pairs are positive, so in this small dataset more study and more sleep both go with higher scores.
Why you cannot compare two covariances
Covariance carries the units of x multiplied by the units of y. If x is measured in hours and y in points, the covariance is in “hour-points” — a unit with no meaning outside that dataset. Change the units (hours to minutes, points to a percentage) and the covariance changes by a large factor even though the relationship is identical. That is the units problem, and it is why a covariance of 59.25 in one dataset and 0.42 in another tells you nothing about which relationship is stronger.
That is exactly why we standardise covariance into Pearson’s r. Dividing by the two standard deviations strips the units away and leaves a number between −1 and 1 that can be compared across any two datasets. Use covariance when you need the raw joint spread (for a regression slope, or for a covariance matrix); use r when you want to compare strength.
The regression connection makes the units concrete. The slope of y on x is Cov(x, y) ÷ Var(x), so the slope inherits the units “y per x”. In Example 1, the slope is 59.2500 ÷ 22.5000 = 2.6333 points of y per unit of x. See the linear regression calculator to fit that line, and the standard deviation calculator for the spread behind it.
Covariance vs correlation
| Covariance | Correlation (r) | |
|---|---|---|
| What it measures | Direction and joint spread of x and y | Direction and strength of the linear relationship |
| Range | Any real number (−∞ to +∞) | Always between −1 and 1 |
| Units | Units of x × units of y | None (unit-free) |
| Comparable across datasets? | No — the size depends on the units | Yes — that is its purpose |
| Use it when | You need a regression slope or a covariance matrix | You want to compare strength or report the relationship |
The two are the same idea at different scales: r is covariance divided by the two standard deviations. The correlation coefficient calculator puts r first, with interpretation and a scatter plot; this page puts covariance first, with r alongside.
Check: the calculator against the test matrix
Run this row through the paired mode and confirm the numbers match: x = [2, 4, 6, 8, 10], y = [9, 7, 5, 4, 2]. The means are x̄ = 6.0000 and ȳ = 5.4000; the cross-products sum to −34.0000. The sample covariance is −34.0000 ÷ 4 = −8.5000, the population covariance is −34.0000 ÷ 5 = −6.8000, and Pearson’s r = −0.9948. A strong negative relationship: as x rises, y falls.
Frequently asked questions
What does covariance tell you?
Covariance tells you the direction of the linear relationship between two variables, and how much they vary together. A positive value means they tend to rise together; a negative value means one tends to fall as the other rises; zero means no linear pattern. Its size also reflects the spread of the data, so it is not a pure measure of strength — that is what the correlation is for.
Why can’t you compare two covariances directly?
Because covariance carries the units of x times the units of y. Two datasets in different units (or with different spreads) will have covariances on completely different scales even when the underlying relationship is identical. Standardise it — divide by the two standard deviations — and you get Pearson’s r, which is unit-free and comparable.
What is the difference between sample and population covariance?
The sample covariance divides the sum of cross-products by n − 1; the population covariance divides by n. Use the sample version when your data are a sample from a larger population (the usual case); use the population version only when your data are the whole population. The sample value is always slightly larger in magnitude.
How is covariance related to correlation?
Correlation is covariance standardised: r = Cov ÷ (SDx × SDy). Equivalently, Cov = r × SDx × SDy. The correlation is always between −1 and 1; covariance has no fixed range. They always have the same sign, and r = 0 exactly when the covariance is 0.
Source: the sample covariance is the standard textbook definition, Cov(x, y) = Σ(xᵢ − x̄)(yᵢ − ȳ) ÷ (n − 1), with the population version dividing by n. The identity Cov = r × SDx × SDy and the regression slope Cov ÷ Var(x) follow directly from that definition.
See all 67 free statistics tools — every calculator on this site in one place.
Next steps
More Tools to Explore
Keep going with the tools that pair with covariance — the correlation, the regression line, the spread and the variance.
Correlation Coefficient Calculator
Pearson’s r — the standardised version of covariance, with interpretation and a scatter plot.
Linear Regression Calculator
Fit the least-squares line — its slope is Cov ÷ Var(x).
Standard Deviation Calculator
The spread behind covariance — work out the standard deviations first.
Standard Deviation vs Variance
Variance vs standard deviation — which one to quote, and when.
Statistics Calculator
Mean, median, mode, range and more for any list of numbers.
Explore All Tools
Browse all 67 free statistics tools—every calculator on this site in one place.