An investigation of incident frequency, duration and lanes blockage for determining traffic delay

Journal of Advanced Transportation, Fall, 2009 by Yi "Grace" Qi, Hualiang "Harry" Teng, David R. Martinelli

Introduction

Traffic congestion can be caused by recurrent events such as the cyclical AM or PM peaks of traffic demand on any given day and by non-recurrent events such as collision or non-collision incidents. Some actions, such as car-pooling and flexible office hours, have been taken under the umbrella of transportation management to reduce and spread the peaks of traffic demand. At the same time, measures have also been applied to enhance incident management with the objective of reducing the traffic impacts caused by incidents which occur in a random manner. In evaluating the traffic congestion impact caused by an incident, the following formula has been widely used. See Wirasinghe (1978) for the derivation:

Delay = [T.sup.2.sub.1]([S.sub.1] - [S.sub.3])([S.sub.2] - [S.sub.3])/2 ([S.sub.1]- [S.sub.2]) (1)

where [T.sub.1] denotes incident duration, [S.sub.1] is capacity, [S.sub.2] represents traffic demand, and [S.sub.3] expresses bottleneck capacity. From this formula it can be seen that the variables that directly related to the traffic delay caused by one incident are [T.sub.1] and [S.sub.3]. Total traffic delay caused by q incidents can be derived by multiplying the delay above for a single incident by the number of incidents that occurred on a roadway. Thus, this simplified method indicates that traffic delay is highly related to three variables: incident frequency (q), incident duration ([T.sub.1]), and the number of lanes blocked by an incident which is directly related to the bottleneck capacity [S.sub.3].

Note that these three variables have been studied to varying extents. Relatively, incident duration has been more extensively studied than incident frequency and the number of lanes blocked in an incident. Most of the studies on frequency were for accidents or a category of accidents. There have been few studies on the frequency of incidents which include both collisions (e.g., property damage, injury, or fatality incidents) and non-collision incidents (e.g., disabled vehicles and unscheduled road repair incidents). The studies on lane blockage in an incident are rare. In this study, we provided an investigation of the influencing factors for all three of these variables based on an incident data set that was collected in New York City (NYC). The information about the incidents derived from the identification can be used by incident management agencies in NYC for strategic policy decision making and daily incident management and traffic operation. It can also be used by interested agencies in other cities to gain a better understanding of incidents.

In this study, incident frequency, incident duration, and the number of lanes blocked in an incident were investigated one after the other. In identifying the influencing factors for incident frequency, a set of models, such as Poisson and negative binomial regression models and their zero-inflated models, were considered. An appropriate model was determined based on a model decision-making tree developed in this study. The influencing factors for incident duration were identified based on hazard-based models where the exponential, Weibull, log-logistic, and log-normal distributions were considered for incident duration. For the number of lanes blocked in an incident, the identification of the influencing factors was based on an ordered probit model which can better capture the order inherent in the number of lanes blocked during an incident.

In the following paper, the first section is devoted to a literature review on the studies of incident frequency, duration, and the number of lanes blocked in an incident. The second section provides a description of the incident data that was collected in NYC and fully used in this study. The third section introduces the models that were considered in the modeling and the ways to choose an appropriate model for each of the following: incident frequency, duration, and the number of lanes blocked in an incident. The fourth section presents the interpretation of the modeling results. The last section provides the conclusions that were derived from the study and the identification of study needs for the future.

Literature Review

In contrast with the abundant literature on the studies of accident frequency, there are few studies on incident frequency. Thus, only the literature on accident frequency was reviewed, aiming to help to establish a methodology for the investigation of incident frequency. In general, accident frequency has been studied from different perspectives and using varied methodologies. For example, it has been studied either for varying levels of severity such as property damage, injury, or fatality, or for the different areas where the accidents happened such as urban versus rural, freeway versus street, intersection versus middle block, etc. Most of the studies focus on identifying the influencing factors of accident frequency. A wide spectrum of models have been employed, such as the linear regression model (Hamerslag et al. 1982), Poisson regression, negative binomial (Hadi et al. 1995, Karlaftis and Tarko 1998), as well as zero-inflated Poisson and negative binomial regression models (Lee and Mannering 1999).