Loading...
Statistical modelling of spatial and temporally referenced data, with applications in finance, ecology, and epidemiology
Date
2025
Abstract
Spatial and spatiotemporal data often violate many underlying assumptions of classical statistical modelling. Many modelling approaches assume that observations are independent, which is not the case for spatially referenced data. To account for the non-independence of observations, or the influence of neighbouring regions or points, a suite of specific modelling techniques can be used. In this research, we consider three spatial and spatiotemporal problems using data collected within the Republic of Ireland.
First, we consider the influence and impact of postcodes on the selling price of residential property prices in County Dublin. Using data collected between January and November 2018, we disentangle the impact of postcode labelling and the spatial location of the property. We study the effect of correcting postcodes and the additional value acquired by postcode mislabelling. We compare our proposed model to various machine learning techniques, such as K-Nearest Neighbours and Random Forests, achieving similar accuracy measures. Ultimately, our proposed model has a greater degree of interpretability compared to many machine learning techniques, which is important for various stakeholders.
We then examine the use of spatial modelling to estimate species populations. For the first time, population estimates of three wild deer species in the Republic of Ireland were obtained. Due to the spatially sparse nature of one data source, we model the three species collectively. This allows for the inclusion of underlying between-species correlations and other inferential advantages. We overcome the issues presented by spatial misalignment in our data sources and data sparsity with our modelling framework. Our results are validated using outcomes from studies conducted in the UK and expert opinion.
Lastly, we assess the spatial and latent spread bovine tuberculosis (bTB) at the herd level in County Kerry. We include a temporally-lagged network structure of the animal trades within County Kerry, the first time this component has been investigated in the context of bTB in Ireland. As disease eradication is the ultimate goal for stakeholders, we examine the one-step-ahead predictions of the proposed model and frameworks proposed in previous studies. While computationally burdensome, our model has many avenues for meaningful impact to aid in understanding the spread and ultimately the eradication of this disease.
Supervisor
Sweeney, James
Description
Publisher
University of Limerick
Citation
Collections
Files
Funding code
Funding Information
Sustainable Development Goals
External Link
Type
Thesis
Rights
http://creativecommons.org/licenses/by-nc-sa/4.0/