Figure 2.

The trade‐off between the expected proportion of patients allocated to the superior arm, $\mathbb{E}\left(p^{*}\right)$, and power for the: binary ER, normal ER, binary FLGI, normal FLGI, and normal FLGI with missing data (MD) imputed in an online fashion for block sizes $b=\left(1,2,4,8,16,32,64,128\right)$ in a two‐armed trial of size $T=128$. The latter two designs are shown when assuming both an unknown variance and known (correct) variance (dashed line and labeled as FLGI‐known). ER, equal randomization; FLGI, Forward‐Looking Gittins Index [This figure appears in color in the electronic version of this article, and any mention of color refers to that version]