Local influence on bandwidth estimation for kernel smoothing. (English) Zbl 1004.62033
Summary: This paper examines the problem of assessing local influence on the optimal bandwidth estimation in kernel smoothing based on cross validation. The bandwidth for kernel smoothing plays an important role in the model fitting and is often estimated using the cross-validation criterion. Following the argument of the second-order approach to local influence suggested by X. Wu and Z. Luo [J. R. Stat. Soc., Ser. B 55, No. 4, 929-936 (1993; Zbl 0798.62032)], we develop a new diagnostic statistic to examine the local influence of the observations on the estimation of the optimal bandwidth, where the perturbation may belong to one of three schemes. These are the response perturbation in the explanatory variable, and the case-weight perturbation.
The proposed diagnostic is nonparametric and is capable of identifying influential observations with strong influence on the bandwidth estimation. An example is presented to illustrate the application of the proposed diagnostic, and the usefulness of the nonparametric approach is illustrated in comparison with some other approaches to the assessment of local influence.
The proposed diagnostic is nonparametric and is capable of identifying influential observations with strong influence on the bandwidth estimation. An example is presented to illustrate the application of the proposed diagnostic, and the usefulness of the nonparametric approach is illustrated in comparison with some other approaches to the assessment of local influence.
MSC:
62G08 | Nonparametric regression and quantile regression |
62J20 | Diagnostics, and linear inference and regression |
62G07 | Density estimation |