|
Abstract:
|
Incorporation of flexible nonlinear functional relationships may be important in several applications. Smoothing splines are frequently used to estimate nonparametric functions and they are particularly useful when complex nonlinear relationships are present and are difficult to be parametrically modeled. Limited literature is available discussing how to choose the most appropriate splines model, which includes natural splines, B-splines and penalized splines. In this work we summarize and compare different splines for modeling nonlinear relationships using linear and logistic regression models. Comparisons and discussion are done through applications and simulation studies. In general, number of knots and spline degree depend on functional form to be modeled, with penalized splines been preferred to natural and B-splines when larger number of knots is required.
|
