areas. Data from such studies thus consist of the number of birds banded each
year and the number of these that are recovered (dead bird encountered by
a member of the public, and the band reported to the investigators, usually via
a central bird-banding repository such as the US Bird Banding Laboratory or the
British Trust for Ornithology).
Such band recovery data are modeled using survival parameters, corresponding
to the probability of a bird surviving from the time of banding in 1 year to the
time of banding in the following year. Sampling parameters analogous to capture
probabilities are also needed in the modeling. In the special case where recoveries
are restricted to reports from hunters of birds that have been shot, recovery rates
(one kind of sampling parameter) are of interest themselves as indices of hunting
intensity (e.g. Anderson 1975; Brownie et al. 1985). Band recovery models
are simply a special case of capture–recapture models. Band recovery models
do not assume that all dead birds are encountered and reported, but instead
view the number of birds reported as some unknown fraction of the total number
dying in a year. A key difference between capture–recapture and band recovery
studies involves the interpretation of the estimated survival parameter. As noted
above, the complement of apparent or local survival in capture–recapture studies
includes both death and permanent emigration. In most band recovery studies
based on recoveries of dead birds by members of the public, most or all dead birds
have some non zero probability of being recovered regardless of where death
occurs. Thus, permanent emigration is not possible and resulting survival estimates
can be viewed as estimates of true survival (complement includes only mortality).
Estimation models for band recovery data were initially developed by
Haldane (1955), Seber (1970), and Robson and Youngs (1971). An excellent
synthetic treatment containing models with different underlying assumptions
about time-specificity of parameters was provided by Brownie et al. (1985).
Survival can be modeled as a function of covariates in band recovery models
(North and Morgan 1979; Conroy et al. 1989). Modeling, model selection, and
estimation are now in most cases best conducted using program MARK (White
and Burnham 1999). Recent discussion of modeling and estimation using
band recovery data is provided by Williams et al. (2002). Capture–recapture data
and band recovery data can be combined for the purpose of estimation. Because
of the different interpretations of survival parameters in the two classes of
models, the combination of methods permits separate estimation of true survival
and fidelity, the probability of returning to the study area (banding location)
conditional on survival (Burnham 1993; Williams et al. 2002).
Study designs for band recovery studies typically involve banding at one
of more local areas at a specific time each year (e.g. at the end of each breeding
season). As with capture–recapture designs, the banding period should be
128 |Estimating survival and movement