The trap of hidden processes: Why 'quick & dirty' methods to estimate mortality are not always good. A comment to De Pascalis et al. (2020)

Conditional analyses, like the one used by De Pascalis et al. (2020), have the risk to overlook important hidden processes. The proportion of animal seen alive vs those seen dead depends, by definition, on the detection probability, p, that an animal alive is detected and on the probability, λ, that a dead animal is found or retrieved (c) and the tag reported (r), with λ = c*r. In CaptureMark-Recapture and -Recovery models (CMRR), p and λ relate the observations to the latent mortality processes governed by the survival probability, S. The estimate of mortality during a given time interval based only on observations makes the implicit assumption that p = λ = 1. However, this assumption is in most cases unrealistic. The danger of a conditional approach when p and λ are different from 1 can be illustrated through simulations.