Testing goodness of fit of polynomial models via spline smoothing techniques

A new test statistic is proposed for testing goodness of fit of an mth order polynomial regression model. The test statistic is [integral operator]10[[mu](m)[lambda](t)]2 dt, where [mu](m)[lambda] is the mth order derivative of a 2mth order smoothing spline estimator for the regression function [mu] and [lambda] is its associated smoothing parameter. The large sample properties of the test statistic are derived under both the null hypothesis and local alternatives. A numerical example is included that illustrates the technique.