Testing for handling bias in survival estimation for black brant

We used an ultrastructure approach in program SURVIV to test for, and remove, bias in survival estimates for the year following mass banding of female black brant (Branta bernicla nigricans). We used relative banding-drive size as the independent variable to control for handling effects in our ultrastructure models, which took the form: S = S0(1 - ??D), where ?? was handling effect and D was the ratio of banding-drive size to the largest banding drive. Brant were divided into 3 classes: goslings, initial captures, and recaptures, based on their state at the time of banding, because we anticipated the potential for heterogeneity in model parameters among classes of brant. Among models examined, for which ?? was not constrained, a model with ?? constant across classes of brant and years, constant survival rates among years for initially captured brant but year-specific survival rates for goslings and recaptures, and year- and class-specific detection probabilities had the lowest Akaike Information Criterion (AIC). Handling effect, ??, was -0.47 ?? 0.13 SE, -0.14 ?? 0.057, and -0.12 ?? 0.049 for goslings, initially released adults, and recaptured adults. Gosling annual survival in the first year ranged from 0.738 ?? 0.072 for the 1986 cohort to 0.260 ?? 0.025 for the 1991 cohort. Inclusion of winter observations increased estimates of first-year survival rates by an average of 30%, suggesting that permanent emigration had an important influence on apparent survival, especially for later cohorts. We estimated annual survival for initially captured brant as 0.782 ?? 0.013, while that for recaptures varied from 0.726 ?? 0.034 to 0.900 ?? 0.062. Our analyses failed to detect a negative effect of handling on survival of brant, which is consistent with an hypothesis of substantial inherent heterogeneity in post-fledging survival rates, such that individuals most likely to die as a result of handling also have lower inherent survival probabilities.