Assessment of biases in the estimation of tag shedding rates through a mark-resight experiment

Abstract

Several flexible methods for estimating tag loss rates through mark-resight data have been developed recently. They allow evaluation of different tag shedding modalities and relax the usual assumption of independence between the loss of tags made in classical double-tagging methods. Two conditions limit the applicability of these methods: (1) tagged individuals must have permanent marks so that they can still be identified after losing their tags, and (2) a large number of observations is required to obtain precise estimates. Here we evaluate the performance of alternative estimators of tag shedding rates when these conditions are not met, a situation that is very common in mark-resight experiments on reef fishes. We simulated resighting data using a simple exponential model of tag shedding under different scenarios created by varying the probabilities of fish detection and fish emigration from the reef, and the tagging schedule. The model was conditioned on actual data from a short-term (∼1.5 years) double-tagging study conducted on the Argentine sandperch Pseudopercis semifasciata (Cuvier 1829) in rocky reefs of northern Patagonia. We tested eight estimation procedures: three variants of an individual-based method, two based on a binomial likelihood function for exact or pooled times-at-liberty data, and three regression methods. Although the individual-based approach produced unbiased and most precise estimates when individuals that had lost both tags were identifiable, it performed poorly in the absence of permanent tags. In contrast, conditional methods, which do not require identification of individuals that have lost both tags, were more robust, providing unbiased and precise estimates. The pros and cons of the different methods for analyzing small-scale mark-resight experiments are discussed.