Statistical Methods

Record the assumptions of a Savitzky–Golay smoothing window

Define the task

A local polynomial smoother estimates each value from a neighboring window. Decide which time scale must be preserved and which high-frequency variation is plausibly noise. A window that spans a real short event can attenuate it. Record the original sampling interval and assess gaps before applying a method that treats neighboring samples as equally spaced.

Carry out the check

Specify window length, polynomial order, derivative order, sample spacing and edge mode. Polynomial order must be less than window length. If estimating a derivative, the spacing determines its units. Edge handling is substantive: polynomial interpolation at the boundary differs from reflection or padding. Compare several scientifically plausible windows and retain the raw signal for inspection; visual smoothness is not an accuracy criterion.

Worked hypothetical example

For hypothetical samples from y=x² at integer x=−2,−1,0,1,2, a quadratic fit over the five-point window reproduces the center value 0 exactly in exact arithmetic. A simple five-point moving average would give 2 instead. This illustrates polynomial preservation, not general immunity to noise. The derivative at the center is 0 and the second derivative is 2 when the spacing is one. Validate software settings with a known polynomial before interpreting measurements, and state how boundary estimates were produced.

Method checked 15 November 2026 against SciPy documentation: savgol_filter.

Related reading

Reading path: Analyze data. Guide 150 of 152. Choose any guide to suit your project.

Statistical Methods

Apply a Hann taper without confusing window gain and signal amplitude

A practical guide with worked examples: Apply a Hann taper and document its gain and endpoint convention.

Statistical Methods

Record endpoint conditions for a cubic spline

Specify endpoint assumptions when interpolating with cubic splines.

Browse this reading pathTools for this stageGuides and downloads for this stage