Plenary Address - 3:50-4:30 PM

Coping with unobservable and misclassified states in multi-state capture-recapture studies

Bill Kendall

Multi-state capture-recapture models have proven to be very useful in the study of animal population and meta-population dynamics. Principal parameters of interest include state-specific survival and transition probabilities. Depending on the study, transition probabilities can refer to movement between sub-populations (e.g., breeding colonies) or transitions between life history stages (e.g., size classes, breeding states). Inherent to traditional multi-state models are the assumptions that in each capture period each animal in the population or meta-population is observable (i.e., subject to detection), and the state of each captured animal is known with certainty. In fact violations of these assumptions are not uncommon. Examples of temporary movement of marked animals out of the study area can include roaming within a home range, movement underground (e.g., transition into torpor), and transition from breeding to non-breeding state where the study areas consist of breeding colonies. Misclassification of the state of an animal can occur when a cue that determines state is missed or misinterpreted. Breeders can be misclassified as non-breeders when young are present but not observed. Diseased animals can be misclassified as healthy because symptoms are absent or not observed. In other cases (e.g., sexually monomorphic bird species) sex can remain undetermined for several or all encounters because sex-specific behavioral cues are not observed.

I will outline the impact of unobservable and misclassified states on estimators for survival and transition, derived from models that ignore these phenomena. I will then describe approaches for coping with biases associated with unobservable and misclassified states. These approaches will include both design (to minimize the nuisance aspect of these phenomena) and modeling (to adjust for these phenomena) elements. The latter will often require the collection of extra information, including capture data from subsamples (i.e., the robust design), telemetry, band recoveries, or ancillary observations.



Individual Papers

4:30-4:50 PM

Costs of first reproduction in a long-lived bird: effects of environmental and individual covariates - Christophe Barbraud & Henri Weimerskirch

How animals balance their investment in young against their own chances to survive and reproduce in the future? This life-history trade-off, referred to as the cost of reproduction, holds a central place in life-history theory. Long-lived birds are good candidates to be used as models to detect these costs. They should be more restrictive than short-lived birds in the degree to which they exhibit increased effort, because even a small reduction in adult survival would reduce the number of subsequent breeding attempts, thereby greatly lowering lifetime reproductive success. However, at least two factors are likely to confound the measurement of this trade-off in the wild. First, there could be differences in the amount of energy individuals acquire and allocate to various functions. In that case we might expect that some individuals would perform well in both reproduction and survival, whereas low quality individuals would die sooner. Second, there could be variations in resource availability affecting energy acquisition and allocation. Theoretical models examining the optimal phenotypic balance between reproduction and adult survival under variable breeding conditions have recently investigated the second issue. However, very little is known on the influence of individual quality on the costs of reproduction. Here, we use a capture-recapture dataset on blue petrels to investigate the costs of first reproduction. The use of multi-state models with three states (non-breeder, first time breeder, and experienced breeder) allowed us to show that first time breeders have a lower probability of breeding the following year than experienced breeders, and that in some years, first time breeders have a lower survival probability that experienced and non-breeders. These results suggest that first time breeding may act as a filter, selecting good quality individuals. Using environmental and individual covariates, we further show that the costs of first reproduction are particularly acute in years when environmental conditions are poor, and that individual body condition affects both survival and breeding probabilities.

4:50-5:10 PM

Density dependence in North American ducks - S.P. Brooks and Lara E. Jamieson

The existence or otherwise of density dependence within a population can have important implications for the management of that population. Here, we use estimates of abundance obtained from annual aerial counts on the major breeding grounds of a variety of North American duck species and use a state space model to separate the observation and ecological system processes. This state space approach allows us to impose a density dependence structure upon the true underlying population rather than on the estimates and we demonstrate the improved robustness of this procedure for detecting density dependence in the population. We also show how the inclusion of time-varying covariates such as the number of May ponds provides additional descriptive power within the model and that their omission may sometimes lead to erroneous conclusions as to the presence of density dependence. We adopt a Bayesian approach to model fitting, using Markov chain Monte Carlo (MCMC) methods and use a reversible jump MCMC scheme to calculate posterior model probabilities which assign probabilities to the presence of density dependence within the population, for example. We show how these probabilities can be used either to discriminate between models or to provide model-averaged predictions which fully account for both parameter and model uncertainty.

5:10-5:30 PM

Demographic estimation methods for plants in the presence of dormancy - Marc Kery & Kathy B. Gregg

Demographic analysis seems straightforward in plants due to their sessile nature. Problems arise when there is an unobservable dormant state that stays belowground for one or more growing seasons. Conventional analysis methods make strong assumptions about the duration of dormancy and obtain estimates of demographic parameters, which, however, will be biased to an unknown degree. In contrast, we use capture-recapture (CR) methods to obtain unbiased estimates of the fraction of a population that is in the dormant state, and of survival and transition rates between life-states. As an illustration, we analyze a 10-year data set on Cleistes bifaria, a terrestrial orchid with frequent dormancy, using both single- and multi-state CR models as well as five conventional methods for comparison. During the study period, 35% of ramets were dormant at least once, for between 1 and 4 (mean 1.4) years. Capture-recapture models estimated ramet survival rate at 0.86 (SE ~0.01), ranging 0.77-0.94 (SE<=0.1 ) in any one year. Average fraction dormant was estimated at 29% (SE 1.5), ranging 16-47% (SE<=5.1) in any one year. Survival rate was positively related to both precipitation in the current year and mean spring temperature. Transition rates were more strongly related to cumulative precipitation in the previous than in the current year: more ramets became dormant following dryer years. Conventional methods augment the number of transitions actually observed by up to 60%. None of them came close to the presumably more unbiased estimates from CR models and, more important, none was consistently best among conventional methods. Formal comparison between conventional and CR methods would ideally involve simulated data. However, our results suggest that CR methods provide less biased estimates under less restrictive assumptions. They should be considered seriously in any demographic study of plant species with dormancy.

5:30-5:50 PM

Determination of sex in Larus Audouinii. A model incorporating a possibility of error - Roger Pradel, Maurin-Bernier, Oro, Olivier Gimenez

In a monomorphic species such as the Audouins gull, determination of sex from behaviour is prone to errors. Repeated observations may allow to attribute a gender to an animal with reasonable confidence. However, when capture events are rare because capture rate is low or because the animal has entered the data set late during the study period, mistakes remain possible. We developped a model where we account for uncertainty in the assessment of sex. This model has more parameters than the corresponding model where the true sex is assumed to be the one most frequently given. We examine whether this causes parameter redundancy when the parameters are constant or time-dependent and we discuss how useful it is to incorporate this additional parameter.

5:50-6:10 PM

Multi-state analysis of the impacts of avian pox on a population of Serins (Serinus serinus) in northeastern Spain. - Juan Carlos Senar & Mike Conroy

Disease is one of the evolutionary forces shaping populations. Recent studies have shown that epidemics like avian pox or mycoplasmosis have affected passerine population dynamics, disproportionately killing males and larger individuals and thus selecting for specific morphotypes. However, few studies have estimated the effects of an epidemic by following individual birds, and most studies have been restricted to the study of variation in population means pre- and post-epidemics. Here we analyze, using multistate models, the development and consequences of an avian pox epidemic affecting a population of Serins (Serinus serinus) in northeastern Spain. The epidemics lasted from the end of July to the end of November, with a maximum prevalence in October. Because avian pox can be diagnosed by direct examination of the birds, we were able to estimate predict transition from the state of being uninfected to the state of being infected in relation to individual covariates. This was related, in turn to the probability of survival, presumably lower for infected than infected birds. We additionally tested, by analyzing the relationship of wing length to survival, for the effects of selection of the epidemics on the wing length of the population. We discuss implications for predictive modeling of disease outbreak, probability of survival of epidemics, and selection for individual characteristics.