Improved Safety Performance Functions for Signalized Intersections.

摘要

For this effort, the research team developed new safety performance functions (SPFs) for signalized intersections in Oregon. The modeling dataset consisted of 964 crashes from a total of 73 intersections that were randomly selected based on the presence of a traffic signal (identified through the crash data records). The SPFs were developed using a Poisson-lognormal Generalized Linear Mixed model framework for total crashes and severe injury crashes (coded as KAB). Three SPFs were developed: 1) an SPF for total crashes, which relies on both major and minor AADTs to predict the expected number of crashes; 2) an SPF for KAB crashes, whose predictions derive from both AADTs as well as from the speed limit on the major road; and (3) a severity model to predict the proportion of KAB crashes to be used in combination with the SPF for total crashes. The research analyses determined that the speed limit variable significantly improved the quality of the SPFs and severity model, and as expected, suggests increasing severity with speed differentials. The models were validated spatially and temporally based on additional sites and using an additional year of data. The models all performed well during the validation; however enhanced models to improve model reliability were developed based on the larger dataset. As part of the model development, this research also explored a variety of rules to identify crashes as intersection-related based on the crash geo-location (including the common 250 feet rule). Crashes were manually classified from the combined data available from the geo-location of crashes, the geometric database, and the various fields in the Oregon crash database. These classifications were then compared to a number of rule options for classifying them as intersection crashes. The analysis revealed that the best performing rule is to use crashes that were geo-located within 300 feet of the centerline intersection at signalized locations plus crashes where the crash report indicates that they were associated with a traffic control device (i.e. traffic signal). Finally, this research effort developed models to estimate to estimate minor road AADT for use in safety analysis where this exposure information is not available. These models were developed from data from 66 intersections with known minor and major AADT volumes and validated with data from another 25 intersections. Significant model variables included major AADT, number of approach lanes, functional class, presence of a two-way left-turn lane, and parallel road AADT.