Goodness-Of-Fit Test for Nonparametric Regression Models: Smoothing Spline ANOVA Models as Example

Area of Science:

• Statistics
• Statistical Modeling

Background:

• Nonparametric regression models offer flexibility by not requiring pre-specified functional forms.
• Existing diagnostic statistics for nonparametric models are less developed compared to parametric approaches.

Purpose of the Study:

• To propose a novel goodness-of-fit test for nonparametric regression models with a linear smoother form.
• To extend this testing framework to smoothing spline ANOVA models.
• To identify two key sources of model lack-of-fit: omitted covariates and inadequate functional form.

Main Methods:

• Derivation of estimated residuals from the nonparametric regression model.
• Assessment of statistical dependence between estimated residuals and covariates using the Hilbert-Schmidt Independence Criterion (HSIC).
• Utilization of the bootstrap method for p-value computation.

Main Results:

• The proposed test effectively detects lack-of-fit when covariates are missing or the model form is inappropriate.
• Simulations demonstrate accurate Type I error control and satisfactory power performance.
• The method was successfully applied to a neonatal mental development dataset.

Conclusions:

• The developed goodness-of-fit test provides a valuable diagnostic tool for nonparametric regression, specifically for smoothing spline ANOVA models.
• The test's ability to detect multiple sources of model misspecification enhances its practical utility.
• This method contributes to improving the reliability and interpretability of nonparametric regression analyses.