Spatio-temporal analysis of head injuries in northwest Iran

Exploring the Robustness of Alternative Cluster Detection and the Threshold Distance Method for Crash Hot Spot Analysis: A Study on Vulnerable Road Users

“…Cáceres (2011) applied Getis-Ord Gi* and Moran's I spatial autocorrelation to the same data to validate hot spot locations [36]. In another study [22], the Global Moran's Index is used to identify the spatial pattern of data such as random, dispersed, or clustered data, and then the Getis-Ord Gi* or Local Moran's Index is used [22]. The most common workflow was the use of Moran's I to determine a threshold distance in the spatial autocorrelation and then apply Getis-Ord Gi* with that threshold distance, followed by KDE.…”

“…An overview of the previous methodology used in the spatial analysis of crash data, including diagrams of the spatial analysis procedure used by researchers, is presented in Figure 2. In Figure 2, workflow 1 was used by [22,33,36], [22,36] then applied workflow 2 and workflow 3 after workflow 1; [15,22,26,34,38,42] employed workflow 3 and workflow 4; [18,19,21] employed workflow 5 and workflow 6 to find critical hot spots; workflow 7 was followed by [43,44] to identify hazardous locations for crashes.…”

“…An overview of the previous methodology used in the spatial analysis of crash data, including diagrams of the spatial analysis procedure used by researchers, is presented in Figure 2. In Figure 2, workflow 1 was used by [22,33,36], [22,36] then applied workflow 2and workflow 3 after workflow 1; [15,22,26,34,38,42] employed workflow 3 and workflow 4; [18,19,21] employed workflow 5 and workflow 6 to find critical hot spots; workflow 7 was followed by [43,44] to identify hazardous locations for crashes. Our literature review did not find any research that applied Moran's I to identify clusters based on network-constrained analysis and then to find a threshold distance that would inform a subsequent network-constrained Getis-Ord Gi* hot spot analysis.…”

“…The Moran's Index ranges between +1 and −1. A value that approaches +1 suggests that there is clustering, a value near 0 suggests that the pattern is randomly dispersed, and a value that approaches −1 suggests that the pattern is dispersion or that there is spatial heterogeneity [21,22,59,60]. The null hypothesis of the statistical test on the z-score is that the data form a spatially random pattern.…”

“…The threshold distance used to estimate spatial autocorrelation plays a key role in identifying clusters. Using Euclidean distances for weighted matrices rather than network distances may incorrectly show clustering when one does not exist, and vice versa [16][17][18][19][20][21][22][23][24][25]. Figure 1 illustrates a scenario that further explains the reason for the differences in threshold distances.…”