ABSTRACT
A road collision is a point or an occurrence on a map. However, in many modeling and prediction techniques, traffic collisions tend to be assigned either to links on the road network or to administrative units for aggregate analysis. This then leads to a fundamental problem experienced by many geographers, which is determining the most appropriate size and shape of the spatial units used for analysis, since this may heavily influence the visual message of mapping and the outcome of statistical tests. There has been considerable debate in recent years concerning the optimal length of basic spatial units for road collision analysis. Largely this is a size and scale issue well known to geographers, that is, the modifiable areal unit problem (MAUP). Often the choice of the level of aggregation is constrained by the format of available data, because collision data and explanatory variables are often collected by different agencies. This chapter seeks to address to the problem of network segmentation and optimal length of basic spatial unit, and the ways collisions are assigned to segments. Besides, methods for assessing spatial autocorrelation have existed for several decades and stem from the work of Moran (1948). An extension of spatial autocorrelation analysis for assessing departures from randomness in regression residuals for flows on a network has been explored in the last two decades. This chapter also seeks to explore the notion of network autocorrelation, and how it can be measured using global and local variations (e.g., local Moran’s I). It will take case studies and examples to expand the theory and statistics.
