Web-based Supplementary Materials for 'Fitting Conditional Survival Models to Meta-Analytic Data by Using a Transformation towards Mixed-Efiects Models'

To apply the method proposed in Section 3.2 we need estimated values for the error variance. To obtain estimates we can rely on an asymptotic representation, proposed by Lo and Singh (1986), decomposing F (k) i (t) − F (k) i (t) as an average of i.i.d. terms and a lower order remainder term rik(t), where F (k) i is the continuous failure time distribution function for subjects in cluster i with xij = k. Let G be the censoring distribution, 1−Hik(s) = {1 − F (k) i (s)}{1 − G(s)} and Hu ik(s) = P (Tij ≤ s, δij = 1|xij = k) = ∫ s 0 1−G(y−)dF (k) i (y). It follows from Lo and Singh (1986) that