J. C. Pardo-Fernández, G. Boente
In this talk, we will consider the problem of testing for the equality of regression curves against general alternatives in a fully nonparametric setting. It is well known that outliers or atypical observations in the samples affect the quality of the classical estimators and inferential procedures, particularly when performing tests. Robust methods are then necessary in order to provide correct estimators and inferences.
We will describe and study a new robust test for comparing regression curves. The test statistic is based on an L2-distance between empirical characteristic functions of residuals. To protect against atypical observations, the residuals are obtained by using robust estimates of the regression functions. The asymptotic distribution of the test statistic is analysed and a small Monte Carlo study is performed to investigate the finite sample behaviour of the proposed test.
Keywords: nonparametric regression, robust estimation, hypothesis testing
Scheduled
GT18 Non-parametric statistics III. Nonparametric inference for regression
June 9, 2022 5:10 PM
A15
