报 告 人： 杭卫强 博士
报告题目：Extended Goodness-of-Fit Tests for Regression Models
Goodness-of-fit tests for the conditional mean and conditional variance have been extensively studied in the literature,which concern whether a postulated model represents the underlying conditional mean or variance of the data adequately. However, little attention has been paid to checking adequacy of the whole model beyond the conditional moments. In this paper, we consider the goodness-of-fit of models that are designed to incorporate all relevant information provided by the covariates to the response.This is particularly relevant to the noise model in causality analysis.Combining an measure of independence and resampling method, we propose a new method for the test. Corresponding asymptotic results are established. We demonstrate numerically the performance of our tests and compare the results with existing relevant methods. Using our tests, we check for existence of possible causal effect of a real data set.