No abstract

For a second order rotatable design, the influence of a n observation a t the design stage is measured by comparing the variances of a predicted response in presence and absence of the observation. Maximum, minimum and averages of the variance are used in the comparison over t h e region of interest. If the factorial and axial points in central composite designs are equidistant from the center, it is then shown that they are equally influential under the proposed measures and vice versa. Two illustrative examples are given to demonstrate that the factorial and axial points may or may not be equally influential and moreover, the influence of points on the prediction variance does depend on the region of interest.

No abstract