Abstract
This paper proposes the new usage of orthogonal array experiments having missing data.All of experiments are not put into practice, and some specified experiments are properly eliminated.
Thus it becomes the design under which the information matrix derived from the design matrix of no futility of parameters has inverse matrix. It is non-orthogonal,however,statistical inference becomes possible as orthogonal array experiments which have missing data.
The interaction effects can be detected, also experiments'size becomes more smaller. It is suitable to the requests of experimenters who would like to reduce the number of experiments. As the results of simulation, when there are interactions, it has confirmed that it can be evaluated substantially by orthogonal array having missing data. But, the power of test in case of mixed orthogonal array varied compared with expected values, because the interactions have partial confounding among the other main effects.Simultaneously,it becomes clearly that the ineffective factors are judged to be significant.
When there are no interactions, every factor is orthogonal to each other in case of mixed orthogonal array, and power of test was slightly higher than orthogonal array having missing data,but both of them give approximately the same results.
