Empirical Comparisons of Planar and Network K-functions in Traffic Accident Analysis

Ripley’s K-function is one of the most effective and most common methods to analyze patterns of clustering and regularity in spatial distributions of point events. However, ordinary K-function analysis assumes a continuous planar space with the Euclidean distance, which could be a serious pitfall when one wants to examine phenomena that are constrained by a transportation network, including traffic accidents, traffic violations, and structural deficiencies. To overcome this problem, a version of the K-function method, named network K-function, has been proposed specifically for use on a network space, where events are located only on network links and distances between points are measured along the shortest path on the network. The purpose of this paper is to investigate differences in the performances of the planar (ordinary) and network K-functions when applied to traffic accident data on a highway network. Analyses are implemented based on Monte Carlo simulation and applied to 1997 traffic accident data in the Buffalo, NY area. Results indicate fundamental differences between the K-function methods resulting from their assumptions about space. Namely, the planar K-function tends to over-detect clustering patterns even in a random pattern on a network, because it compares the observed point pattern with a random pattern in a planar space. For instance, a spatial distribution of accidents during weekends, which the network Kfunction revealed to be random, was still identified as clustering by the planar K-function. TRB 2003 Annual Meeting CD-ROM Paper revised from original submittal.