Date of Award
Doctor of Philosophy (PhD)
Nicholas J. Gotelli
Bayesian hierarchical detection models are useful for addressing uncertainty in datasets in the form of detection error and can be adapted to a variety of research questions. This dissertation uses three case studies to highlight advantages of Bayesian hierarchical detection models: 1) using prior information to model undetected species, 2) efficiently modeling a naturally hierarchical system, and 3) correcting for observation bias in two interconnected ecological metrics for effective disease management.Detection error can bias ecological observations, especially when a species is never detected during sampling. In many communities, the probable identity of these species is known from previous research, but these data are rarely included in subsequent models. I present prior aggregation as a method to add information from external sources to Bayesian hierarchical detection models. Prior aggregation combines information from multiple prior distributions: in this case, an ecologically informative, species-level prior and an uninformative community-level prior. This approach adds external information into the model while retaining the advantage of modeling species in the context of the community. Using simulated data supplied to a multi-species occupancy model (MSOM), I demonstrated that prior aggregation improves estimates of metacommunity richness and environmental correlates of species occupancy. When applied to a dataset of Vermont small mammals, prior aggregation allowed the model to estimate occupancy correlates of the eastern cottontail, a species observed at several study sites but never captured. Ectoparasites are exposed to a ‘dual’ environment: the individual host and the external environment. However, variation in the portion of the life cycle spent on-host leads to differences in selective pressures exerted by each environment. Parasites that spend most of the life cycle on-host face increased pressure to specialize, leading to differences in host specificity and occupancy patterns compared to ephemeral parasites which only contact the host to feed. Using data from small mammals and ectoparasites in Vermont, I used a multi-scale MSOM to 1) calculate the Bayesian R2 at the site and host levels of the model to quantify explained variation in occupancy, and 2) compare number of host species and R2 values across life history categories. Life history was significantly associated with host specificity and host-level R2: parasites which spend more time on-host infested fewer hosts and had more variation explained by host traits than ephemeral parasites. However, there were no differences in site-level R2 between categories, suggesting additional factors structure small mammal/ectoparasite communities. Disease management requires accurate measurements of metrics such as population size and immunity rates. Raccoon rabies virus is managed through use of oral rabies vaccine bait distribution, and the efficacy of the strategy is evaluated by measuring population-level seroprevalence of rabies antibodies. Using data from the Burlington, VT area from 2015–2017, I modified a multinomial N-mixture model to 1) estimate raccoon abundance and seroprevalence while correcting for sampling error, and 2) evaluate the effects of management strategies, raccoon population characteristics, and other carnivore species on seroprevalence. Rabies seroprevalence was associated with traits of raccoon populations, increasing with average age and decreasing with population size. Seroprevalence also decreased with opossum captures, suggesting competition for baits. Management strategies did not affect seroprevalence within sampling sites, but there is evidence that baiting strategy affects seroprevalence at the regional level.
Number of Pages
Beasley, Emily, "Applications Of Bayesian Hierarchical Detection Models" (2023). Graduate College Dissertations and Theses. 1670.
Available for download on Sunday, April 07, 2024
Ecology and Evolutionary Biology Commons, Parasitology Commons, Statistics and Probability Commons