A non-parametric sequential rank-sum probability ratio test method for binary hypothesis testing

Abstract

Sequential probability ratio test (SPRT) is a common technique for binary hypothesis testing. To use SPRT, one needs to assume a distribution function, such as Gaussian, for samples. However, in some applications the sample distribution is unknown, non-Gaussian, and/or cannot be specified by a simple function. This paper presents a non-parametric sequential rank-sum probability ratio test (SRPRT) method to conduct binary hypothesis test under such circumstances. The new method only assumes that the sample distribution is symmetric, which is easy to satisfy. Statistical tests are conducted to compare the performance of this non-parametric method to those of SPRT and its variant. Results show that it is more appropriate to use SRPRT than SPRT in binary hypothesis testing if the sample distribution is non-Gaussian.