×

A comparison of optimum design algorithms for regressions with correlated observations – a computational study. (English) Zbl 0988.62044

Two iterative algorithms for the computation of optimum experimental designs are compared numerically on 5 examples with known solutions. The experiments are supposed to be with a given covariance structure, and without replications of observations. Both algorithms contain heuristic steps and are briefly discussed. For the original papers cf. U.N. Brimkulov, G.K. Krug and V.L. Savanov [Numerical construction of exact experimental designs when measurements are correlated. Zavodskaya Laboratoria 36, 435-442 (Russian) (1980)], and W.G. Müller and A. Pázman [An algorithm for the computation of optimum designs under a given covariance structure. Comput. Stat. 14, 197-211 (1999)].
The comparison is done by the frequency of failures in 10 000 runs of the algorithms, each one with a different initial design chosen at random.

MSC:

62K05 Optimal statistical designs
65C60 Computational problems in statistics (MSC2010)
PDFBibTeX XMLCite