Introduction

Abstract

Suppose two treatments are available for a certain disease. Patients arrive at a clinic one at a time and one of the treatments must be used on each. Information as to the effectiveness of the treatments accrues as they are used. The overall objective is to treat as many patients as effectively as possible. This seemingly innocent but important problem is surprisingly difficult, even when the responses are dichotomous, either success or failure. It is an example of a two-armed bandit problem.