Résumé
For comparative demography studies, 2 prerequisites are usually needed: 1) using typical parameter values for species, 2) correctly accounting for the uncertainty in the species specific estimates. However, although within-species variability may be essential, it is typically not considered in analytical procedures, resulting in parameter estimates that may not be representative of the species. Further, data are analysed in 2 steps, first separately for each species, then estimates are compared among species. Accounting for the uncertainty in the species specific estimates is then difficult. Here we propose the application of multilevel Bayesian models on markrecapture (MR) data for comparative studies on survival probabilities that solves these problems. Our models account for within-species variability in space and time in the form of random effects. Models reflecting different biological predictions related to the species ecology and life-history traits may further be contrasted. To illustrate our approach, we used long-term data from 5 temperate tree-roosting bat species and compared their survival probabilities. Results suggest that species foraging in open space, high reproductive output and short longevity records have lower survival than species foraging at short distances, with low reproductive output and high longevity records. Multilevel models provided relatively precise estimates, away from the edges of the parameter space, even for species with low encounter rates and short study duration. This is particularly valuable for less studied taxa such as bats for which available data are often more sparse. Our approach can be easily extended to include additional groups or levels of interest and effects at the individual level (e.g. sex or age). Different hypotheses regarding differences or similarities in parameters among species can be tested through the application of different models. Overall, it offers a flexible tool to ecologists, and population and evolutionary biologists for comparative studies, explicitly accounting for multilevel structures often encountered in MR data.