Evaluation of Hotspots Identification Using Kernel Density Estimation (K) and Getis-Ord (Gi*) on I-630

Abstract

The main objective of this paper is to compare two statistical techniques kernel density estimation (K) and Getis-Ord Gi* statistic using a Geographic Information System (GIS) for hotspot identification. The standardized Gi* is essentially a Z-value associated with statistical significance. The two statistical techniques were compared using seven years of crash data (2000-2006) on I-630 (7.4 miles) in Arkansas. The highway had very high rate of crashes; 457 crashes per mile were observed during the analysis period. I-630 is located in only one county and, therefore, assumed that the demographic effect will be minimal and mainly local spatial autocorrelation will be observed. Results indicated that the estimation by K and Gi* including three conceptualization of spatial relationships (CSR) for hotspots (high; categorized as high and low) were almost similar. Additionally, the three CSR methods (fixed distance, inverse distance, and inverse square distance) identified the same hotspots (high). Also, the range of Z values for Gi* for the hotspots (high) were similar for the three CSR‘s for years 2000 and 2001. For 2002 and 2003, range of Z values for Gi* for the hotspots (high) were similar for inverse distance and inverse square distance CSR. Another data set, aggregated from 2004 to 2006 was used to identify hotspots by K and Gi* (inverse square distance, CSR). Results indicated similar hotspots identified by both methods. The reason may be due to the mathematical functions of K and G*i . Some of the key contributing factors in this paper are also discussed.