Skip to main content

Grades 9-12Free, no account needed

Linear regression calculator

Fit an exact least-squares line to typed point pairs and inspect correlation, r-squared, predictions, residuals and independent normal-equation checks.

All calculators

Try your problem

2-20 pairs separated by semicolons. Example: -1,1/2;0,2;3,4. x must vary. Press Ctrl+Enter to calculate.

Your calculations stay in this browser. No input history is saved.

Supported inputs and limits

Enter 2-20 x,y pairs separated by semicolons, at most 1100 characters: 0,1;1,2;2,2. Each coordinate is a signed integer, decimal with 1-6 places or integer fraction with positive denominator; at most 24 characters and 12 digits per numerator/denominator part, magnitude at most 1000000 and reduced denominator at most 1000000. No expressions, mixed numbers, repeated decimals or exponents. x must vary. Constant y is allowed but r and centered r-squared are undefined. Exact rational means, slope, intercept, residuals, SSE and r-squared; signed radical r is exact, with 10-significant-digit approximation when irrational. Unweighted OLS with intercept only; no nonlinear fit, significance test or prediction interval. Paste regress(0,1;1,2;2,2), case-sensitive and at most 200 characters. Plot coordinates are approximate and tiny residuals are not enlarged.

Understand the method

Points (0,1), (1,2) and (2,2) surround the least-squares line y = 7/6 + x/2. Vertical residuals are -1/6, 1/3 and -1/6, sum to zero, and have squared sum 1/6; r-squared is 3/4.

Ordinary least squares with an intercept chooses y-hat = a + bx to minimize the sum of vertical squared residuals. A residual is observed y minus predicted y; it is not the perpendicular distance to the line.

Center coordinates about their means. Sum centered x squares for Sxx and centered cross-products for Sxy. The slope is Sxy/Sxx, and the intercept is mean y minus slope times mean x. The fitted line passes through the coordinate means.

Pearson r is Sxy divided by the square root of Sxx times Syy. Its sign follows the slope, while r-squared loses the sign. With varying y, r-squared = 1 - SSE/SST. Constant x cannot determine a unique slope; constant y gives a horizontal fit but undefined correlation and centered r-squared. Association alone does not establish causation.

Worked examples

Read the reasoning, or load an example into the calculator.

A fit with nonzero residuals

For (0,1),(1,2),(2,2), Sxx=2 and Sxy=1. Slope=1/2, intercept=7/6, SSE=1/6 and r-squared=3/4. Correlation is sqrt(3/4), approximately 0.8660254038.

Perfect negative association

Points (0,4),(1,2),(2,0) lie on y=4-2x. Residuals vanish, r=-1 and r-squared=1; squaring removes the negative sign.

Constant response

For (0,5),(1,5),(2,5), the exact fit is y=5 with SSE=0. Syy=0, so correlation and centered r-squared are undefined, not 1.

Try it without the calculator

Three questions for practice, not a graded school assessment. Your answers stay on this page and are not saved.

1. A perfect decreasing line has which r and r-squared?
2. For OLS with an intercept, what is the residual sum?
3. Why are equal x coordinates rejected?

Original Testinar learning content. This local preview is awaiting final review before publication.

Page UP