Abstract

A linear network is a combination of line segments, or edges, that run between their defined endpoints, or nodes. They have become increasingly prevalent within spatial statistics due to the potential for representing systems from various fields as linear networks. One specific area of study within linear networks is understanding how they interact with one another and whether spatial similarity correlates with any underlying causal relationships. This line of research, however, remains limited due to the lack of a robust spatial similarity tests suited for linear networks. This research therefore develops a new linear network spatial similarity test that specifically takes into account the spatial context of two linear networks and allows for spatially dependent variations in similarity. Different characteristics of the new test are demonstrated in two separate simulation studies. The first simulation study tests the overall performance. The second simulation study shows the benefit of the newly proposed test compared to an alternative method. Finally, the test is applied to real-world informal road and mobility networks across Northwestern Namibia to test whether mobility routes in rural areas are similar to existing infrastructure, and how the degree of similarity varies across regions. Sub-analyses are also conducted to investigate the effect of road conditions, seasons and road density on the spatial similarity between the two networks.