Listy Biometryczne - Biometrical Letters Vol. 35(1998), No. 2, 117-125
|
DETERMINATION OF THE SIZE OF AN EXPERIMENT IN THE ANALYSIS OF VARIANCE FOR FIXED AND MIXED MODELS* Dieter Rasch and Minghui Wang Sub-department of Mathematics, Department AMST, Wageningen Agricultural University, Dreijenlaan 4, 6703HA Wageningen, The Netherlands e-mail: Dieter.Rasch@wts.wk.wau.nl, Minghui.Wang@wts.wk.wau.nl |
An algorithm and a program for the determination of the minimal size of experiments to compare levels of a fixed factor for balanced classifications in the analysis of variance is described and demonstrated. The minimal size is determined in dependence on a lower bound for the difference between the maximum and the minimum of the effects to be tested for equality by an F-test, on the risks of the test and on a presumable value of the common residual variance. We determine the minimal size for the least favorable case (the maxi-min size) as well as for the most favorable case (the mini-min size). The classification and the model chosen influence the degrees of freedom for the F-test and the relationship between the size of the experiment and the non-centrality parameter.
analysis of variance, minimal size, non-centrality parameter, sub-class number
