Abstract
Motivated by problems arising in the study of road traffic accidents, this thesis develops statistical methodology for analysing spatial patterns of points along a network of lines. Weaknesses of the traditional "crash-frequency" approach to traffic accident data are identified, and we develop a new approach using Poisson point process models. A new, statistically-principled, rapidly computable method of kernel density estimation on a linear network based on the heat kernel is developed. Techniques for estimating relative risk for spatial point patterns in two-dimensional space are extended to point patterns on a linear network and we develop improved methods for bandwidth selection.
Original language | English |
---|---|
Qualification | Doctor of Philosophy |
Awarding Institution |
|
Supervisors/Advisors |
|
Thesis sponsors | |
Award date | 25 May 2020 |
DOIs | |
Publication status | Unpublished - 2019 |