抄録
In this paper, we study the conditional sample numbers given that the null hypothesis has been accepted or rejected in the sequential probability ratio test (SPRT) by attributes. First we derive the moments of conditional sample numbers through formulating the SPRT with the aid of Markov chain. Then we clarify some property of the relative defference between accepted one and rejected one concerning average sample number, and illustrate a sampling inspection plan in which the property plays an important role. In addition since our approach is supposed to give almost exact solutions of the OC and ASN functions as well, we evaluate the approximation errors in the case of Wald's SPRT and further consider how to design a sampling plan which realizes given α and β exactly.
