Answers are grounded in this paper and cite their sources.
Functional semi-parametric regression models are widely used to analyze curve-valued predictors, yet model misspecification can severely bias inference. Existing goodness-of-fit tests either target fully parametric models or ignore the hybrid structure of semi-parametric function-on-scalar regressions. This paper proposes a suite of test statistics that decompose model checking into three components: a functional-part statistic, a parametric-part statistic, and a full covariate statistic, all built upon weighted L² distances between empirical characteristic functions. The functional covariate is first compressed via functional principal component analysis and then combined with scalar covariates; the resulting statistics are formulated as U-statistics with analytically convenient kernels derived from spherical stable or Laplace densities. We rigorously study the asymptotic behavior under both correct specification and a broad class of alternatives. A key theoretical contribution is the proof that, under the null hypothesis, the estimation effect introduced by replacing unobservable true errors with estimated residuals is asymptotically negligible when the permutation procedure is used for calibration; under alternatives the effect is of the same order as the statistic itself and is rigorously quantified. We also provide jackknife variance estimators that are consistent under alternatives, enabling the construction of critical regions without relying on resampling when Δ>0. Monte Carlo simulations demonstrate that the functional-part statistic excels at detecting omitted nonlinearities in the functional component, the parametric-part statistic is sensitive to misspecification of the scalar covariates, and the full statistic captures interactive misspecifications, thereby enabling directional diagnosis of model inadequacy. The tests control size accurately even with heavy-tailed errors and maintain high power across a range of sample sizes and dimensions. A real-data application using near-infrared absorbance spectra of meat samples demonstrates how the decomposed statistics can diagnose different sources of model inadequacy in predicting fat content. The decomposed testing framework provides a practical tool for diagnosing model misspecification in functional semi-parametric regression and can be extended to other complex functional models.
No takes yet. Share an insight, caveat, or question.
Mengru You (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: