An adaptive biased coin design for the behrens-fisher problem

Abstract

Wei(1978) introduced the adaptive biased coin design to reduce experimenter bias and offer a compromise between perfect balance and complete randomization. In situations such as the Behrens-Fisher problem, balance is not necessarily desired and the optimal ratio of sample sizes is unknown. To reduce experimenter bias, by introducing randomization, an adaptive biased coin design is superimposed on Robbins, Simons, and Starr's(1967) sequential analogue of the Behrens-Fisher problem. The design has asymptotic properties similar to Robbins, Simons, and Starr's sequential procedure.

Keywords:

• asymptotic optimality
• experimenter bias
• sampling plan
• sequential design
• stopping rule