Complex observation processes in ecology and epidemiology : general theory and specific examples

• Main Author: Chadwick, Fergus J
• Format: Dissertation
• Language: English
• Published: University of Glasgow 2023
• DOI: 10.5525/gla.thesis.83512

Complex observation processes abound in ecology and epidemiology. In order to answer the large-scale, urgent questions that are the focus of modern research, we must rely on indirect and opportunistic observation. Relating these data to the biological processes we are interested in is challenging. Statisticians working in this area need an understanding of both state-of-the-art modelling techniques and the field-specific nuances of how the data were generated. As a result, many methods to deal with complex observation processes are highly bespoke. Bespoke models are hard to translate between contexts and, because they are often presented in field-specific language, hard to learn from. Modelling of observation processes is thus a fractured area of study, leading to duplication of research effort and limiting the rate at which we can make progress. In this thesis, I aim to provide a road-map to how we might achieve some unification in this area. I begin by establishing a conceptual framework that can be used to describe observation processes and identify methods to address them. The framework defines all observation processes as a combination of issues of latency, identifiability, effort or scaling (L.I.E.S.). I illustrate the framework using motivating examples from ecology and epidemiology. The risk with conceptual frameworks is that they can be over-fitted to existing data and may fail when faced with new, real-world problems. To address this, I also approach the problem from a bottom-up perspective by tackling a series of ecological and epidemiological case studies. Each case study requires novel statistical methods to deal with the observation process. By developing new methods, I explore the world of observation processes potentially not well-captured in the literature. I then explore whether these case studies motivate revision or reassessment of my conceptual framework. While the case studies were chosen to challenge the L.I.E.S. framework, I find that they mutually reinforce each other. The framework provides a helpful scaffolding with which to describe the problems in the case studies. The case studies provide useful examples of more complex observation processes and how the four issues encoded in L.I.E.S. interact with one another. These findings illustrate the value of a framework for unifying approaches to observation processes.