The initial conditions problem in modelling repeated durations: A simulation study

Simulation results and an analysis of real data demonstrate the substantial parameters estimation biases which can occur in analysing repeated duration data if proper account is not taken of interrupted, left-censored, durations at the beginning of the data window. Conditional or joint likelihood methods overcome the initial conditions problem. Simulation studies show that the problem of initial conditions in sampling within a fixed observational window is more serious than in sampling a fixed number of renewals.