Hierarchical Bayes for structured, variable populations: From recapture data to life-history prediction

Understanding population dynamics requires models that admit the complexity of natural populations and the data ecologists obtain from them. Populations possess structure, which may be defined as "fixed" stages through which individuals pass, with superimposed variability among individuals and groups. Data contain missing values and inaccurate censuses. From limited data ecologists attempt to predict life history schedules and population growth. We extend the "missing value" framework for Bayesian analysis of structured populations to admit the heterogeneity in demography and the limitations of data that are typical of ecological populations. Our hierarchical treatment of capture-recapture data allows inference on demographic rates contained in matrix transition models for populations that may be stratified by location and by other variables. Simulations with artificial data sets demonstrate the ability of the Bayesian model to Successfully estimate underlying parameters, even with incomplete census data. In contrast, traditional nonhierarchical models may lead to biased parameter estimates because of variation in recapture rates of individuals. Analyses of published demographic data on Common Terns and Taitu Hills rats illustrate the utility of the model. Predictive distributions of maturation age, survivorship, and population growth demonstrate profound impacts of population and data complexity.