Fits up to 9 standard curves (Linear, Logarithmic, Inverse, Quadratic, Cubic, Power, Compound, S-curve, Exponential) to a single (X, Y) and ranks them by AIC.
A curve-estimation wizard. For exploratory shape-finding when you don't have a parametric form in mind, this fits the standard candidate library and lets you pick the best fit by AIC.
Critical detail: AIC is only directly comparable across models with the same response transform. Log-y fits (Power, Compound, S-curve, Exponential) have a different scale than plain-y fits (Linear, Quadratic, Cubic). We add the Box-Cox correction (+ 2·Σlog(y)) to log-y AICs, putting all 9 models on a comparable scale in the AIC* column.
Numeric X + Y.
Per-model R², adjusted R², σ, AIC, AIC* (Box-Cox corrected), F, p, coefficients. Best model highlighted.
Differences in AIC* below 2 are negligible. Pick the simplest model in the 'effectively-tied' set.