Department of Statistics, University of Waikato
University of Waikato, Room I1.09
When perturbing a regression model, to identify outliers and influential observations, most computer packages use updating procedures to find the inverse of the information matrix. Consider a linear model partitioned into two components, one of which is perturbed and the other left unchanged. In the talk we shall examine how the updating procedure can be extended to the inverse of the information matrix of the perturbed component. We shall also briefly discuss a three component model where the third component changes as the first component is perturbed. The results have important applications in the construction of efficient experimental designs.