Evaluating sources of censoring and truncation in telemetry‐based survival data

Abstract

ABSTRACTBias in vital rate estimation may come from failing to meet a variety of assumptions during the stages of sampling, monitoring, and analysis, though most are not commonly addressed in published studies. Here, we pay specific attention to various forms of censoring and truncation that present challenges for telemetry‐based monitoring of survival. We use simulations to assess how uncertainty about times of death and imperfect detection probabilities affect Kaplan–Meier survival estimates. We then treat monitoring of threatened woodland caribou (Rangifer tarandus caribou) in west‐central Alberta as a case study to test for potential effects of non‐random right censoring and interval censoring on survival estimates. We report that monitoring frequency (e.g., daily vs. monthly) and associated uncertainty about the exact time of death do not inherently induce bias on Kaplan–Meier point estimates of survival nor affect estimates of variance. Removing individuals from the at‐risk pool during intervals for which they were not detected did induce a negative bias on resulting survival estimates when the probability of detection was independent of the animals' fates. We recommend using subsequent detections to impute animals' fates during missed intervals in such cases when all animals' fates are eventually known or permanently right censored and when missed detections are not related to changes in mortality risk. Although some assumptions remain difficult to test and of continued concern, we find no evidence of biases in the methodology of Alberta's woodland caribou monitoring program. Our results lend credence to recent evidence of widespread declines in woodland caribou populations across the province. © 2015 The Wildlife Society.