Accident Analysis and Prevention 40 (2008) 76–88
Modeling left-turn crash occurrence at signalized intersections by conflicting patterns
Department of Civil & Environmental Engineering, University of Central Florida,
Orlando, FL 32816-2450, United States
Received 14 October 2006; received in revised form 26 February 2007; accepted 20 April 2007
Abstract
In order to better understand the underlying crash mechanisms, left-turn crashes occurring at 197 four-legged signalized intersections over 6years were classified into nine patterns based on vehicle maneuvers and then were assigned to intersection approaches. Crash frequency of eachpatternwas modeled at the approach level by mainly using Generalized Estimating Equations (GEE) with the 留學(xué)生dissertation網(wǎng)Negative Binomial as the link functionto account for the correlation among the crash data. GEE with a binomial logit link function was also applied for patterns with fewer crashes.The Cumulative Residuals test shows that, for correlated left-turn crashes, GEE models usually outperformed basic Negative Binomial models.The estimation results show that there are obvious differences in the factors that cause the occurrence of different left-turn collision patterns. Forexample, for each pattern, the traffic flows to which the colliding vehicles belong are identified to be significant. The width of the crossing distance(represented by the number of through lanes on the opposing approach of the left-turning traffic) is associated with more left-turn traffic collidingwith opposing through traffic (Pattern 5), but with less left-turning traffic colliding with near-side crossing through traffic (Pattern 8). The safetyeffectiveness of the left-turning signal is not consistent for different crash patterns; “protected” phasing is correlated with fewer Pattern 5 crashes,but with more Pattern 8 crashes. The study indicates that in order to develop efficient countermeasures for left-turn crashes and improve safety atsignalized intersections, left-turn crashes should be considered in different patterns.
Keywords: Signalized intersection; Left-turn crashes; Conflicting patterns; Approach level model; Negative Binomial; Generalized Estimating Equations
1. Introduction
Analyzing left-turning traffic is crucial for improving intersectionoperation and safety. This traffic may collide with manyother traffic flows at signalized intersections, e.g., with leftturningtraffic from the same approach or from the differentapproaches, and with through traffic from other approaches;therefore, left-turn crashes have many distinct conflicting patternsin vehicle maneuvers before collisions. Left-turn crashesare among the most frequently occurring collision types; basedon the crash history of 1531 signalized intersections in the stateof Florida, left-turn crashes rank third, following rear-end andangle crashes, and represent 16% of all intersection reportedcrashes (Abdel-Aty et al., 2005).At signalized intersections, differences exist in traffic volumes,site geometry, and signal operations, as well as safety∗ Corresponding author. Tel.: +1 407 8235657; fax: +1 407 8234676.E-mail address: [email protected] (M. Abdel-Aty).#p#分頁(yè)標(biāo)題#e#
performance on various approaches of intersections. Therefore,modeling the total number of left-turn crashes at intersectionsmay obscure the real relationship between the crash causes
(i.e., intersection characteristics, etc.) and their effects (i.e., leftturncrashes, etc.). However, at signalized intersections, trafficflows and signal operations of different approaches are interactive
at intersections; therefore, disaggregating intersectionsinto approaches will introduce correlation among observations
rom the same intersection. In this study, left-turn crashes werenvestigated at the approach level by conflicting patterns basedon geometry and traffic-related explanatory variables using the
appropriate statistical models, which are able to analyze correlatedcrash frequencies.
1.1. Factors affecting left-turn crash occurrence
Several studies have attempted to quantify the effects oftraffic flow, intersection geometric design features, and trafficcontrol and operational features on left-turn crash occurrence.0001-4575/$ – see front matter © 2007 Elsevier Ltd. All rights reserved.doi:10.1016/j.aap.2007.04.006
X. Wang, M. Abdel-Aty / Accident Analysis and Prevention 40 (2008) 76–88 77Poch and Mannering (1996) fitted an approach level left-turncrash frequency model and identified that left-turning movement,opposing approach volume, type of traffic control, type ofleft-turn signal, speed limit, and sight distance all have an effecton left-turn crashes. Pernia et al. (2002) identified that overall
traffic, number of lanes on a major road, and the presence ofa median are significant. Hauer et al. (1988) assumed that the
frequency of collisions is related to the traffic flows to whichthe colliding vehicles belong and not to the sum of the enteringflows.On signalized intersection approaches, left-turning trafficcan be treated in one of three ways: “permissive”, “compound”(“protected/permissive” or “permissive/protected”), and“protected”,1 and numerous studies have been conducted forevaluating their safety effect. Agent (1987) found where “permissive”phasing was replaced by “protected/permissive”, thenumber of left-turn crashes usually decreased, except on oneapproach, where the speed limit was larger than 45 mph.
Upchurch (1991) compared crash rates (left-turn crashes/millioneft-turning vehicles) for different left-turn phasing and foundthat “compound” signal has a higher crash rate than “permissive”phasing. Many researchers have reached the conclusionthat “protected/permissive” phasing has more left-turn crashesthan “protected” phasing in before-after studies (Benioff and
Rorabaugh, 1980; Warren, 1985). Lee et al. (1991) found thatthere was no significant difference in crash experience betweenleading and lagging operations.
It has been found that as the width of a median increases, thesight distance for left-turning vehicles decreases significantly(Harwood et al., 1996; Yan and Radwan, 2004). For unprotectedleft-turn traffic at signalized intersections, vehicles turning leftfrom opposing left-turn lanes often restrict each other’s sightdistance (Joshua and Saka, 1992; McCoy et al., 1992). Joshuaand Saka (1992) pointed out that the minimum value of the offsetcan be zero, but cannot be a negative value in practical design,which would results in unsafe conditions.#p#分頁(yè)標(biāo)題#e#
1.2. Crash modeling strategyPernia et al. (2002) fitted an intersection level model forleft-turn crashes, but they found that the model explaining thecrash variation is very limited. Poch and Mannering (1996) fittedan approach level left-turn crash frequency model, of whichcrashes were assigned to the approach of the “at-fault” vehicle.Following this criterion, if a northbound left-turning vehicle collidedwith a southbound through vehicle, and if the left-turningvehicle/driver was at-fault, this crash would be assigned to thenorthbound approach. The signs of some variables are contraryto the expectation, e.g., the opposing left-turn volume has a negativecoefficient. Hauer et al. (1988) classified crashes into 15patterns based on vehicle maneuvers before the collisions; Pat-
1 For “permissive”, left-turning traffic is permitted to cross through the opposingflow. For “protected”, left-turning traffic is protected by stopping theopposing through traffic. More complicated is “compound”, which includes“protected/permissive” or “permissive/protected” depending on the order of thesequence.terns 5 through 12 involve a left-turning vehicle, and then crasheswere assigned to the approach near the crash site. Separate crashmodels were fitted at the approach level for each pattern based
only on conflicting flows.Since at signalized intersections traffic flows and signal operationsof different approaches are interactive, disaggregatingan intersection into four approaches may produce correlationamong the data. There are serious problems arising when basiccount data models are used for correlated data, since basic countdata models assume the dependent variables are independent.For a model at the approach level, the data structures becomea mixture of randomly selected intersections and highly correlatedapproaches within an intersection. Generalized EstimatingEquations (GEE) provide an extension of generalized linearmodels (GLMs) to the analysis of correlated data, which canaccount for the correlation among the data. In addition, sincecrashes are rare events, disaggregating left-turn crashes by collisionpattern and by approach may lead to a high proportion ofzeros and ones in some collision patterns. GEE with a binomiallogit link function can be attempted for the patterns with fewer
crashes.
1.3. Research objective
The primary purpose of this study is to investigate the relationshipbetween left-turn crash occurrence and intersection
features, i.e., geometry design features, traffic control and operationalfeatures, traffic flows, etc. The left-turn crashes were
divided into different conflicting patterns based on the vehiclemaneuvers before collisions, and then assigned to the approachwith left-turning vehicles. GEEs, which provide an extension of
generalized linear models (GLMs) to analyze correlated data,were applied for separate left-turn crash models to account forthe correlation among the data.
2. Data preparation#p#分頁(yè)標(biāo)題#e#
Information on intersection geometry design features, trafficcontrol and operational features, traffic flows, and crashes
from 2000 to 2005 were obtained for 197 four-legged signalizedintersections from Orange and Hillsborough counties inthe Central Florida area. The intersections with at least onestate road were used since the data for these intersections arewell-maintained. The intersections which had major changes ingeometry, signal timing, and traffic flow during the study periodwere excluded; therefore, the traffic operations for the selectedintersections remained fairly uniform during the study period.The selected intersections are on level sites.
2.1. Intersection characteristics
Geometry design features for the intersections were extractedby inspecting the aerial imagery contained in the software
http://www.ttc-switch.com/Thesis_Writing/Accounting_Assignment/Engineering/Google Earth (2005), which puts high-resolution aerial and satellite
imagery and other geographic information on the desktop.
The number of through lanes on each approach, the number
of left turn lanes and whether they were exclusive, the pres