Abstract
This paper deals with statistical comparison between several kinds of signal-response systems having the same objective function.Assume that these systems have a common binary signal and a quantitative response, respectively. The problem is formulated by the statistical test of the homogeneity of the noncentrality parameters of noncentral t distributions. Consider the case where the same set of test items is assigned to the several systems so that observed quantitative observations become dependent. New simple testing procedures are derived based on the variance stabilizing transformation. It is shown that the proposed procedures are conservative in almost every case. The influence of dependencies on the efficiencies is also investigated.It is shown that assigning the same set to different systems is almost more efficient than using independent sets.
