BOTTOM-UP NETWORK SCREENING TO IDENTIFY HIGH COLLISION LOCATIONS FOR THE CITY OF SASKATOON

Abstract

Safety network screening is used to identify roadway locations (e.g., intersections and roadway segments) for potential safety improvements. Currently, one of the most commonly used network screening methods in practice is the safety performance function (SPF) based method that uses traffic volume data as an essential input for the screening process. However, the lack of traffic volume data for target roadway locations restricts the applicability of SPF-based network screening methods. The primary objective of this study is to screen Saskatoon’s roadway networks using two existing network screening methods (i.e., the binomial test and the beta-binomial (BB) test) that do not require traffic volume as an input. Previous studies have applied the binomial test and/or the BB test without explicitly defining the particular circumstances that indicate which test is preferable. This study introduced a formal statistical test known as the overdispersion test (i.e., “C(α) Test”) to determine which network screening method – the binomial test or the BB test – should be used to screen a given study dataset. The “C(α) Test” was applied to a total of 36 study collision datasets, including 26 segment collision datasets, and 10 intersection collision datasets. (“C (α) Test” results showed that 15 of 26 (58%) segment collision datasets, and all of 10 intersection collision datasets contained statistically significant overdispersion at the 95% confidence level (P-value < 0.05). The BB test was selected as an appropriate network screening method for 15 segment collision datasets and 10 intersection collision datasets. The remaining 11 segment collision datasets that did not contain statistically significant overdispersion (P-value ≥ 0.05) were screened using the binomial test. The network screening results for each study location (i.e., a segment or an intersection) in all 36 study datasets were presented in terms of the estimated probability obtained from either the binomial test or the BB test. The estimated probability values were used as a ranking measure to select the top 10 or top 30 riskiest locations for both roadway segments and intersections. The network screening results (estimated probability) for each study segment or intersection in all 36 study collision datasets were then visually displayed in a set of 36 collision maps that were developed using ArcGIS. The developed GIS-based collision maps are expected to help engineers in the City of Saskatoon to efficiently select potential locations for deploying specific safety countermeasures that will result in the reduction of a certain configuration of collisions at the screened locations. As a final component of this thesis, a diagnosis study was performed to identify the most dominant collision configurations at the top 30 riskiest signalized intersections (among a total of 154 signalized intersections) in Saskatoon. This study quantitatively compared the performance of two existing collision diagnosis methods (i.e., descriptive data analysis and BB test), and the comparison results revealed that the BB test is a more rigorous collision diagnosis method than the descriptive data analysis.