Figure 3.

The trade‐offs between the expected proportion of patients allocated to the superior arm, $\mathbb{E}\left(p^{*}\right)$, power, average absolute bias in the treatment effect estimate and variability of patient allocations for the different designs, including normal FLGI and normal FLGI with missing data (MD), for block sizes $b=\left(1,15,40,60\right)$ in a three‐armed trial of size $T=120$ (assuming unknown variance). CFLGI, Controlled FLGI; ER, Equal Randomization; FLGI, Forward‐Looking Gittins Index; TP, Trippa Procedure; TS, Thompson Sampling [This figure appears in color in the electronic version of this article, and any mention of color refers to that version]