48
For Xj > xi, we have
(3.17)
if subject i is censored; (extended from Equation 3.6).
(3.18)
if subject i is failed; (extended from Equation 3.8).
Using the above computation results and other specifications as described earlier,
replace the unknown functions by their estimates at the initial step, and then max-
imize the likelihood functions in Equations 3.10 and 3.11 with respect to β and βc,
respectively. Denote the resulting estimators for β and βc by β(v> and βc'* ■
Step 2,
Use β^,βim∖P^m~1∖β,Qfim-1∖β, Dim~1∖β and E∫m"υ(∙) to obtain S⅛n∖β and
⅛m,(<) = elp{-A<">(()}
i -≥h
∑xi<x,∙ <5xi,2^7^υ(¾)
-⅛m 1,(≈3)e≈p(⅛∕3(m))
∑.j≤√'"-1>⅛)
(3.19)
Xj<t