Honors Thesis: Efficient Spatial and Temporal Generalized Linear Modeling

Spatiotemporal data is a common occurrence in a variety of fields such as ecology and epidemiology. However, observations are often spatially and temporally correlated, leading naive models of such data to underestimate variance in parameter estimates. Furthermore, methods that do account for this spatial and temporal dependence often take prohibitively long to estimate due to their computational complexity. This paper extends a computationally efficient method for spatial modeling to the spatiotemporal domain while retaining its computational efficiency. We implement this method and examine its effectiveness using a simulation study and by applying it to Carolina Wren population counts in the United States between 1990 and 2010. We find that it performs favorably compared to the naive approach and is significantly more computationally efficient compared to the full spatiotemporal model. Additionally, it requires much less expert knowledge to specify compared to comparable methods, making this method an attractive approach for users with less experience with spatiotemporal modeling.

My full honors paper can be found here. Slides for my honors presentation can be found here.