A new approach to influence analysis in linear models

Abstract

propose a new approach to the study of influence in the General Linear Model based on conditional bias. This approach enables us to apply such an analysis to all particular cases of this model. The theoretical foundation, on which this approach is based, does not presuppose a particular hypothesis on the distribution of the variables. Applying the results obtained to the Multiple Linear Regression Model, measures of influence are obtained as already proposed by other authors. Finally we carry out an application of the results on the analysis of covariance.