Disease dynamics are often strongly influenced by random variation, or stochasticity, which can be due to either environmental fluctuations or the chance events that befall individuals. The random nature of individual life histories, so-called “demographic stochasticity”, has effects commonly thought to dissipate in the large populations that support epidemics. Epidemiological modeling therefore largely focuses on incorporating “environmental stochasticity”. However, limited dispersal can cause large populations to behave as locally-small. Uneven mixing could potentially enhance the effects of demographic stochasticity on epidemic severity, even at high population densities, challenging the relative importance of environmental stochasticity for spatially-structured systems.

To investigate the importance of demographic stochasticity in disease transmission, we collected spatial data on infection rates of a species-specific baculovirus in Douglas-fir tussock moth larvae, Orgyia pseudotsugata. We sampled branch-level infection rates during O. pseudotsugata outbreaks that spanned several orders of magnitude in population size. We built agent-based models, which incorporate demographic stochasticity, and random differential equation models, which incorporate environmental stochasticity. We then fully parameterized each spatial model to our data using an efficient statistical routine. We tested whether the data for each population were better described by environmental versus demographic stochasticity using the LOO model selection criterion.

Our model selection analysis definitively demonstrates that models incorporating demographic stochasticity are better able to describe the spatial distribution of infection load over time for the O. pseudotsugata system. This effect occurs because limited movement and chance differences in individual fates lead to strongly divergent disease dynamics between different trees. Our study reveals persistent effects of demographic stochasticity across a broad range of population densities, and suggests that models of environmental stochasticity may be of limited usefulness for understanding animal and plant pathogens. We’ll be presenting this work in a forthcoming paper soon!