Challenges in evaluating avian survival
Adult survival estimates are affected by several methodological caveats that we consider here. First, a general problem with comparing survival studies is that differences between estimates derived from old versus new methods and between live recaptures and dead recoveries may mask trends in the data (Roodbergen et al. 2012). Our dataset consisted primarily of studies that used live capture-mark-recapture techniques (83% of effect sizes) and most of these were conducted since 2000; nearly all studies were conducted after 1990 when modern statistical tools for analyses of marked animals were developed (Lebreton et al. 1992). One of the drawbacks of capture-mark-recapture data is that the reported metric, apparent survival, is a product of true survival and site fidelity and as such will always be biased low, whereas estimates of survival from dead recovery models are often interpreted as true survival (Sandercock 2006). Biases in survival estimates may therefore be strong for birds from tropical regions, which consisted exclusively of live-recapture data, and where behaviors such as altitudinal migration are more common than in temperate regions (Barçante et al. 2017) and can lead to temporary emigration from study plots. Another issue affecting the comparison of survival studies is the study duration. This, too, may be particularly problematic for tropical regions, where data collection is often hampered by sampling conducted over irregular or insufficiently long intervals to produce robust estimates of survival (Ruiz-Guitérrezet al. 2012). For example, in our meta-analysis 62% and 69% of effect sizes from austral and temperate latitudes, respectively, were calculated from datasets spanning >10 years, compared to only 34% from tropical latitudes. However, in a study of tropical birds comparing survival estimates derived from 6 vs.12 years of data, Blake & Loiselle (2013) reported an improvement in precision, but no change in point estimates for survival. Still, other authors argue that longer time frames are needed to generate reliable survival estimates for tropical resident species (i.e., 10–30 years), given their expected longevity and low recapture probabilities (p <0.25; Ruiz-Guitérrez et al. 2012). Despite these problems with the comparability of the data, we found no indication that difference in methodological approaches strongly biased our results (Fig. S3).