How to fit one 01 Paste the pairs , one x, y line each; commas or spaces separate the two numbers. 02 Click a family. The equation card and the three coefficient cards update, and the plot redraws. 03 Set the prediction x and read ŷ from the accent card. 04 Take it onward. The equation exports as a math.js expression for the data workspace , and the coefficient table lists each fitted parameter.
Worked examples Three pastes, three different jobs for the same panel:
A symmetric U. 1, 10, 2, 5, 3, 2, 4, 1, 5, 2, 6, 5, 7, 10 fitted as Quadratic reads y = x^2 − 8x + 17 with R² = 1 ; ŷ at x = 8 is 17 . On the same points, the linear card shows R² = 0 , and both Pearson r and Spearman ρ read 0 .
Growth data. 0, 2, 1, 2.7, 2, 3.6, 3, 4.9, 4, 6.6, 5, 8.9 fitted as Exponential reads y = 1.9968·e^(0.29869x) , R² = 0.99997 ; the linear alternative is y = 1.3571x + 1.3905 , R² = 0.95507 . At x = 6 the exponential predicts 11.9856 against the line's 9.53333 .
A power law. 1, 2, 2, 8, 3, 18, 4, 32, 5, 50 fitted as Power reads y = 2·x^2 with R² = 1 , while Pearson r shows only 0.9811 — the data are exactly quadratic in x, not linear.
Limits to keep in mind
Fit only, no inference. No parameter standard errors, p-values, confidence intervals or prediction bands; ŷ is a point.
Domain guards drop rows. Logarithmic needs x > 0, exponential needs y > 0, power needs both positive — out-of-domain pairs are left out, so a fit can rest on fewer points than you pasted. Sinusoidal needs at least five points; logistic needs y > 0. What it does not do. One predictor, one fit at a time: no multiple regression, no interaction terms, no residual diagnostics, no model comparison table. For time-series extrapolation with a rolling backtest use Forecast Lab ; for summaries and hypothesis tests of the same data, see Descriptive Statistics , Tests and the statistics workbench .
A perfect curve can read r = 0 In the U-shaped example the quadratic R² is 1 — the curve passes through all seven points — yet Pearson r and Spearman ρ both read 0 . They are not contradicting R²: Pearson r only measures linear co-variation, and this cloud is symmetric about its center; Spearman ρ collapses because every y value's rank ties in the middle. The reading rule is simple: r answers "how straight is it", R² answers "how well does the chosen curve explain y". A near-zero r with visible structure in the plot is a prompt to switch family, not a finding of "no relationship".
Where it helps Choosing between a line and a curve For the growth data the exponential R² of 0.99997 beats the linear 0.95507 , and the predicted value at x = 6 differs by more than two units (11.9856 vs 9.53333 ). Compare candidate families on the same paste before writing any number down.
Turning a fit into a reusable expression The equation card exports a math.js string — the power fit above exports 2*x^2, the quadratic exports 17 - 8*x + 1*x^2 — which the data workspace stores as a variable and the scientific calculator can evaluate at a new x.
Privacy Pairs are fitted and plotted in the browser; nothing is uploaded and the equation leaves the page only when you copy or send it.
References
Wikipedia, Polynomial regression , en.wikipedia.org (访问日期:2026-10-01)— the least-squares polynomials behind the four polynomial families.
Wikipedia, Coefficient of determination , en.wikipedia.org (访问日期:2026-10-01)— what R² does and does not measure.
Wikipedia, Pearson correlation coefficient , en.wikipedia.org (访问日期:2026-10-01)— linear co-variation only.
Wikipedia, Spearman's rank correlation coefficient , en.wikipedia.org (访问日期:2026-10-01)— rank correlation and what ties do to it.
Wikipedia, Curve fitting , en.wikipedia.org (访问日期:2026-10-01)— the fit-and-predict framing.
Calculators in this hub
Hand-picked tools, one click away. The mini versions compute live and carry your values into the full calculator.
Sources & review
Reviewed by CalcX Editorial Team
Updated 2026-10-01