Figure 5.

Coverage probabilities and bias of mean and variance estimates for the case study. The first row of panels shows actual-nominal coverage probabilities (AC-NC) for the 97.5% upper (dashed line), lower (dotted line) and the 95% two-sided confidence intervals, for the unadjusted reassessment rule $n_2^u$. The second row shows the bias of the mean estimate if $n_2^u$ (solid line) or $n_2^a$ (dashed line) is used. The dotted line shows upper and lower bounds for the bias for general blinded sample size reassessment rules based on $S_{1,\mathit{OS}}^2$. The dashed gray line shows the bounds for the bias for a general unblinded sample size reassessment rule. The third panel shows the bias of the variance estimate if either $n_2^u$ (solid line) or $n_2^a$ (dashed line) is used. The red line shows the theoretical boundary for the bias give in Theorem 4.