Journal of Statistics: Advances in Theory and ApplicationsAuthor's: F. I. Nsude and P. E. Chigbu
Pages: [75] - [97]
Received Date: August 5, 2014
Submitted by:
The method presented here is a variance gradient based approach for constructing exact D-optimal designs. Variance of regression function is approximated by a polynomial function of degree 2n′. Maximizer of this polynomial is determined by a variance modulated line search. Thereafter, this maximizer is used to replace the point of minimum prediction variance amongst the points already in the design. This exchange of points is continued until the sequence converges to D-optimal design. This method has a rapid convergence and compares with the method reported by Atkinson and Donev. Finally, the computer execution time of this method is highly encouraged.
gradient vector, variance function, convex combination, D-optimal, experimental space.
