Residual diagnostics for assessing closed population capture–recapture models

Abstract Capture–recapture models provide a statistical framework for estimating demographic parameters from incomplete observation data, where not all individuals in a population are detected during sampling. Assessing the fit of such models is crucial for reliable inference. However, capture–recapture models often involve non‐standard probability distributions, making standard goodness‐of‐fit (GOF) tools difficult to apply and frequently requiring bespoke programming. To address this challenge, we adopt a partial likelihood approach that conditions on the initial capture of individuals and explicitly models the subsequent recapture process . This formulation recasts the problem as a binomial generalized linear model for post‐entry capture events, thereby enabling the use of the widely adopted R ‐package DHARMa for simulation‐based GOF assessment of the recapture component in closed population capture–recapture models. We provide practical guidance for implementing the partial likelihood approach to assess GOF in capture–recapture models. The diagnostic capabilities of the DHARMa package are illustrated using both simulated and real ecological datasets. Our results show that DHARMa , within the partial likelihood framework, effectively detects misspecification in the recapture process , including heterogeneity, temporal variation, and Markovian behavioural (trap‐response) effects in (re)capture probabilities. Furthermore, diagnostic power increases with the amount of recapture information, indicating that studies with either extensive sampling effort or high capture probabilities are particularly well suited to this framework.