Robust inference for randomized play the winner design

This dissertation develops statistical methodologies for analysis of design parameters and response variables in a randomized play the winner design (RPWD). RPWD is an allocation procedure that changes the allocation probability during the experiment so as to allocate more subjects to a better performing treatment. This causes an imbalance in the number of subjects allocated to various treatment arms. Standard statistical methodologies that do not account for the adaptiveness in the allocation lead to procedures that are inefficient and inaccurate. In this thesis, several novel methodologies using bootstrap and Hellinger distance are developed to mitigate the problems of adaptive allocation. These methodologies have potential for application in other ares of science and technology.