Fitting Curves

Nonparametric Fits

SAS/INSIGHT software provides nonparametric curve-fitting estimates from smoothing spline, kernel, loess, and fixed bandwidth local polynomial estimators that are alternatives to fitting polynomials. Because nonparametric methods allow more flexibility for the functional dependence of Y on X than a typical parametric model does, nonparametric methods are well suited for situations where little is known about the process under study.

To carry out a nonparametric regression, you need first to determine the smoothness of the fit. With SAS/INSIGHT software, you can specify a particular value for a smoothing parameter, specify a particular degrees of freedom for a smoother, or request a default best fit. The data are then smoothed to estimate the regression curve. This is in contrast to the parametric regression where the degree of the polynomial controls the complexity of the fit. For the polynomial, additional complexity can result in inappropriate global behavior. Nonparametric methods allow local use of additional complexity and thus are better tools to capture complex behavior than polynomials.

Normal Kernel Fit

Loess Smoothing

Top of Page