Résumé
A recent trend in the analysis of capture-recapture data, in the broad sense, is the development of survival models mixing different types of events such as live recapture and dead recoveries. These models are similar to those developed for analysing competing events and multistate life histories in human health studies, which are frequently based on Markov chains. The major difference is that, in the capture-recapture context, every event is subject to a detection probability in general lower than one. Multistratum capture-recapture models are then a natural tool. We show that models for information mixtures easily enter this framework, provided adequate, possibly non-observable, strata are defined. Similarities and differences between such models are then easy to ascertain and understand. A multistratum model for mixtures of live recaptures and dead recoveries is presented and an example of data from the literature is treated, using a prototype MATLAB program named M-SURGE. Further generalizations and expected developments are discussed.