Analysing point patterns on networks — A review

Abstract

We review recent research on statistical methods for analysing spatial patterns of points on a network of lines, such as road accident locations along a road network. Due to geometrical complexities, the analysis of such data is extremely challenging, and we describe several common methodological errors. The intrinsic lack of homogeneity in a network militates against the traditional methods of spatial statistics based on stationary processes. Topics include kernel density estimation, relative risk estimation, parametric and non-parametric modelling of intensity, second-order analysis using the K-function and pair correlation function, and point process model construction. An important message is that the choice of distance metric on the network is pivotal in the theoretical development and in the analysis of real data. Challenges for statistical computation are discussed and open-source software is provided.