A Markov chain model for geographical accessibility

Abstract

Accessibility analyses are conducted for a variety of applications, including urban planning and public health studies. These applications may aggregate data at the level of administrative units, such as provinces or municipalities. Accessibility between administrative units can be quantified by travel distance. However, modelling the distances between all administrative units in a region is computationally expensive if a large number of administrative units is considered. We propose a methodology to model accessibility between administrative units as a homogeneous Markov chain, where the administrative units are states and standardised inverse travel distances act as transition probabilities. Single transitions are allowed only between adjacent administrative units, resulting in a sparse one-step transition probability matrix (TPM). Powers of the TPM are taken to obtain transition probabilities between non-adjacent units. The methodology assumes that the Markov property holds for travel between units. We apply the methodology to administrative units within Tshwane, South Africa, considering only major roads for the sake of computation. The results are compared to those obtained using Euclidean distance, and we show that using network distance yields more reasonable results. The proposed methodology is computationally efficient and can be used to estimate accessibility between any set of administrative units connected by a road network.