The most common duration models consider that there is only one event of interest and that this event can only occur once for each individual (i.e., two-state models or one-way transition models). An important assumption for this kind of models is that survival times are independent. Nevertheless, more complicated situations exist that involve multiple events (i.e., more than one event, of the same or different type, can occur to a given individual). In this situation, the assumption of independent survival times is probably not satisfied (Box-Steffensmeier and Jones, 2004; Cleves, 1999; Kleinbaum and Klein, 2005), because the several survival times for the same individual are probably correlated (the second and posterior events are probably to be affected by the previous events). If this correlation is not considered in the model, the estimates of the coefficients of the covariates are probably biased and the variance estimates could be misleading (Aalen, 1992), because the amount of information about each observation is overstated (Box-Steffensmeier and Zorn, 2002). As such, this paper intends to present some alternative ways to model duration models in presence of multiple events.

--

Duration models,multiple event models