Semiparametric Regression with Cubic Spline

Systematic errors contained in observations are always complicated smooth function varying with some variables. This paper describes this systematic errors using natural cubic spline, which is nonparametric component in semiparametric regression model. Penalised least squares technique implemented in the procedure reduces to unique solution. According to simulating tests, the semiparametric regression model and the penalised least squares technique can better separate systematic errors from observations compared with the parametric model and the least squares technique.