Mark loss can strongly bias demographic rates in multi-state models: a case study with simulated and empirical datasets

Abstract

The development of methods for individual identification in wild species and the refinement of Capture-Mark-Recapture (CMR) models during the past decades has greatly improved the assessment of population demographic rates to answer ecological and conservation questions. In particular, multistate models, with their flexibility for the analysis of complex study systems, have become popular in the ecological community. However, despite the extensive use of these models, little attention has been paid to the effect of common violations of the CMR model assumptions, such as mark loss and the often-associated recycling of remarked individuals. To explore this knowledge gap we used a wide range of simulation scenarios reflecting frequently encountered real case studies inspired from the survival rates of 700 vertebrates' species. We estimated the effects of mark loss and recycled individuals on parameter estimates using a multi-state Cormack-Jolly-Seber (MSCJS) framework. We explored parameter bias through simulations of a metapopulation system with different capture and survival rates. We also illustrated how mark loss can be easily estimated and accounted for using an empirical long-term (10 years) CMR dataset of bats, individually identified using both PIT tag technology as marks that can be lost, and multi-locus genotypes as "permanent marks". The results from our simulated scenarios demonstrated that the occurrence of bias and the parameters concerned were highly dependent on the study system, and no general rules on parameter behaviour can be established a priori. The model structure and the interdependency among parameters make it challenging to predict how bias could affect estimates. Our results highlight the need to assess the effect of mark loss when using MSCJS models. Ignoring such violations of model assumptions can have important implications for ecological inferences and conservation policies. In general, the use of permanent marks (e.g. genotype), should always be preferred when modelling of population dynamics and if not possible, combining two types of temporary marks (e.g. PIT tags, bands) should be considered.

Generation and applications of simulated datasets to integrate social network and demographic analyses

“…To compare the success of models that used network measures calculated from the partial versus full network we calculated the earth mover's distance (EMD) of the posterior distributions (Touzalin et al, 2022) for the parameter of interest from relevant pairs of models.…”

Wing Tattoos: A Cost-Effective and Long-Lasting Method for Marking Bats

“…Banding, however, have been used successfully and without apparent injuries, especially on neonatal bats (Rossiter et al ., 2002), and does provide highly valuable information on bat populations and structure. PIT tagging does not cause injuries, but tags are frequently lost (Rigby et al ., 2012), and tag failures or losses can lead to incorrect estimation of survival rates, and an overestimation of population size (McDonald et al ., 2003; Seber and Schofield, 2019; Touzalin et al ., 2022). Loss of PIT tags can be reduced by the application of tissue adhesive (Touzalin et al ., 2022), but not completely eliminated.…”

“…PIT tagging does not cause injuries, but tags are frequently lost (Rigby et al ., 2012), and tag failures or losses can lead to incorrect estimation of survival rates, and an overestimation of population size (McDonald et al ., 2003; Seber and Schofield, 2019; Touzalin et al ., 2022). Loss of PIT tags can be reduced by the application of tissue adhesive (Touzalin et al ., 2022), but not completely eliminated. Tag loss of 2.7% have been reported in bats that were marked with both PIT tags and forearm bands (Van Harten et al ., 2021); however, survival rate of bats was not estimated and as such no measure of the number of bats that could possibly have died as a result of injuries from forearm bands was provided.…”

Generation and applications of simulated datasets to integrate social network and demographic analyses

“…To compare the success of models that used network measures calculated from the partial versus full network, we calculated the earth mover's distance (EMD) of the posterior distributions (Touzalin et al, 2022 ) for the parameter of interest from relevant pairs of models. (i.e., we calculated the EMD for model versions using the full and partial network together with longitudinal imputation and also the EMD for the model versions using the full and partial network together with cross‐sectional imputation).…”