Alternative Bayesian analysis of capture recapture data with behavioral effect modelling

n the context of capture-recapture experiments we consider a generalized version of a model framework recently proposed in Chao and Yang (2005) to better understand behavioural pattern in response to trapping experience. A general order Markov structure allows to incorporate in the analysis flexible behavioural effects on the capture probabilities. We point out that the conditional likelihood approach used in the original work to carry out inference yields unbounded estimates with positive probability. We derive conditions under which pathological inference occurs and connect it to a similar problem highlighted within a different restricted removal sampling framework. To overcome the likelihood failure we investigate alternative Bayesian estimators under different non-informative prior distributions and verify with a simulation study their comparative merits in terms of efficiency and interval estimate coverage.