Fig. 1.
Optimal EF or ESL assignments of patient M = 50 in a clinical trial with horizon M h = 100 as a function of t and w 0 (Left) or w 1 (Right). The value t measures the certainty with which one treatment is known to be superior, whereas w 0 or w 1 parameterizes the imbalance of sample sizes. Blue indicates assignment to A, red to B. The rather sharp boundary in (t,w 0) coordinates is even sharper, and more linear, in (t,w 1) coordinates. A nearly optimal heuristic strategy is to assign the apparently superior treatment (by t-value) unless it is too overrepresented in the sample (by w 1 value). The green line indicates the divide.