On Improving Telemetry-Based Survival Estimation

Abstract Survival estimation is an important aspect of population ecology and conservation biology, and radiotelemetry is a major tool for assessing factors influencing survival time in free-ranging birds and mammals. Despite the advantage of telemetry in providing extensive and continuous survival information, telemetry-based survival estimates can be biased or imprecise when methods are misused. Simple cumulative survival estimators like the Heisey and Fuller and Kaplan-Meier methods have underlying assumptions and sampling requirements that commonly remain unverified by researchers. Telemetry studies often limit survival analysis to simple univariate tests that do not consider the range of factors potentially influencing mortality risk in free-ranging animals. Continuous-time modeling approaches like Cox Proportional Hazards or Anderson–Gill methods, or their discrete-time analogues, are superior because they are robust to a range of study design limitations and can handle multiple categorical or continuous covariates including those that vary with time or subject age. Parametric models may be difficult to fit in telemetry studies because the appropriate hazard function in wildlife populations usually is not known. The main assumptions in survival study design are that 1) subjects represent the population of interest, 2) mortality risk is independent between subjects, and 3) subjects are lost to follow-up (i.e., censored) randomly. These assumptions are prone to violation in telemetry research, and their assessment and possible remediation should be prioritized. Telemetry studies often are characterized by small sample size or short duration; both attributes lead to low numbers of mortalities and thus lack of precision in the survival estimate. I conclude that telemetry-based survival estimation will benefit from increased emphasis on modeling approaches designed to elucidate survival determinants in complex systems, combined with more rigorous attention to basic assumptions and study design limitations.