Fig. 2From: Augmenting the logrank test in the design of clinical trials in which non-proportional hazards of the treatment effect may be anticipatedSmoothed scatter plots of scaled Schoenfeld residuals for the treatment variable in the three trials denoted by (a), (b) and (c). Grey shaded areas show pointwise 95 % confidence intervals. Solid horizontal lines show y= ln HR, the (constant) log hazard ratio estimated by a Cox model. Dashed horizontal lines show y=0, corresponding to ln(HR) =0, null treatment effectBack to article page