(1964)。另一方面有時僅是因資料中所存 在的一個觀測值或數個觀測值，則必須考

The assumption of normality provides the customary powerful and convenient way of analyzing linear regression problem and multivariate data. The problem of non-normality may often be simplified by an appropriate transformation, e.g. the parametric family of power transformations of Box and Cox (1964). The evidence for transformations may sometimes depend crucially on e one or a few observations. Therefore, multivariate data transformations are very sensitive to outliers. The purpose of the paper is to develop methods that would not be influenced by potential outliers during the process of data transformations. They essentially need robust statistics. We propose a robust likelihood ratio test for the transformation parameters. The resulting methods will be able to verify the role of every observation playing in the data transformation as well as to provide the robust estimation after transformation