Commercial Software

Easy-to use programs are available to estimateregression surfaces using nonparametric regression (a type of nonlinear regression, in which the shape of the surface is not constrained to conform to a particular parametric equation). Univariate surfaces are estimated using the cubic spline. Multivariate surfaces are estimated using an additive approximation with univariate surfaces ("projection pursuit regression"). X-variables must be continuous. Y-variables can be normal, binomial (0 or 1), or Poisson (>=0). Bootstrap standard errors of the surface are provided. Free programs are available as FORTRAN source code, or compiled programs for the PC and Sun.
Contact schluter@bcu.ubc.ca.