Effects of individual misidentification on estimates of survival in long-term mark–resight studies

Abstract

All ecological measurements are subject to error; the effects of missed detection (false negatives) are well known, but the effects of mistaken detection (false positives) are less understood. Long-term capture–recapture datasets provide valuable ecological insights and baselines for conservation and management, but where such studies rely on noninvasive re-encounters, such as field-readable color bands, there is the potential to accumulate detection errors as the length of the study and number of tags deployed increases. We investigated the prevalence and effects of misreads in a 10-yr dataset of Red Knots ( Calidris canutus rufa ) marked with field-readable leg flags in Delaware, USA. We quantified the effects of misreads on survival estimation via a simulation study and evaluated whether removal of individuals only reported once in a year (potential misreads) influenced survival estimation from both simulated datasets and our case study data. We found overall apparent error rates of 0.31% (minimum) to 6.6% (maximum). Observer-specific error rates and the variation among observers both decreased with the number of flags an observer recorded. Our simulation study showed that misreads lead to spurious negative trends in survival over time, particularly for long-term studies. Removing all records in which a flag was only recorded once in a sampling occasion reduced bias and eliminated spurious negative trends in survival but also reduced precision in survival estimates. Without data filtering, we found a slight decrease in Red Knot annual survival probability from 2008 to 2018 (β = −0.043 ± 0.03), but removing all single-observation records resulted in no apparent trend (β = −0.0074 ± 0.02). Spurious trends in demographic rates could influence inference about population trajectories and resultant conservation decision-making. Data filtering could eliminate errors, but researchers should carefully consider the tradeoff between precision obtained by larger sample sizes and potential bias due to misreads in their data.