Let each subject contribute observation time to
the appropriate group according to covariate at the relevant
time
•Proportional
hazards model
–Easily
done, if noninformative censoring results
•Accelerated
failure time model
–Difficult due to need to integrate hazards over disjoint
intervals