Approximate sequential test for the Behrens-Fisher problem

AbstractA sequential test for the Behrens-Fisher problem is considered in this paper. A sequential probability ratio test (SPRT), based on sequence of approximate t statistics, suitable for practical use is proposed and its properties are investigated by a simulation study. The test is found to be conservative in terms of desired risk levels. Notwithstanding, the test appears to be quite satisfactory in terms of the power and the average sample number (ASN) function, and it is found to be sufficiently more efficient compared to the approximate fixed-sample test. Also studied by the simulation study is the approximate SPRT for slightly different forms of the hypotheses given by Ghosh[6]. The results suggest that the risk levels of the test tend to exceed the fixed levels