The Variable Selection in Covariance Adjusted Estimates

In this paper we study the problem about how to select the covariables. A new test approach by using the canonical correlation is proposed, the approximate distribution of the test statistics is gived. A simulation comparision is made under three forms of covariance matrices which are adopted very often in the economy, biology and medicine. Our results show that the necessity of covariable selection and the good properity of the canonical correlation test method.