Partial vs. Full Likelihoods


New Member

When fitting a proportional hazards model with time-dependent covariates for the transition intensities of a Markov chain, it seems that using a full likelihood can lead to non-zero sensitivities to covariates for a transition that does not have any events, while this is not possible using partial likelihoods. Is my understanding correct? If so, can anyone suggest a sensible remedy?

Thanks in advance.