Sudeshna Mitra

On the significance of omitted variables in intersection crash modeling

Advances in safety research--trying to improve the collective understanding of motor vehicle crash causes and contributing factors--rest upon the pursuit of numerous lines of research inquiry. The research community has focused considerable attention on analytical methods development (negative binomial models, simultaneous equations, etc.), on better experimental designs (before-after studies, comparison sites, etc.), on improving exposure measures, and on model specification improvements (additive terms, non-linear relations, etc.). One might logically seek to know which lines of inquiry might provide the most significant improvements in understanding crash causation and/or prediction. It is the contention of this paper that the exclusion of important variables (causal or surrogate measures of causal variables) cause omitted variable bias in model estimation and is an important and neglected line of inquiry in safety research. In particular, spatially related variables are often difficult to collect and omitted from crash models--but offer significant opportunities to better understand contributing factors and/or causes of crashes. This study examines the role of important variables (other than Average Annual Daily Traffic (AADT)) that are generally omitted from intersection crash prediction models. In addition to the geometric and traffic regulatory information of intersection, the proposed model includes many spatial factors such as local influences of weather, sun glare, proximity to drinking establishments, and proximity to schools--representing a mix of potential environmental and human factors that are theoretically important, but rarely used. Results suggest that these variables in addition to AADT have significant explanatory power, and their exclusion leads to omitted variable bias. Provided is evidence that variable exclusion overstates the effect of minor road AADT by as much as 40% and major road AADT by 14%.

Spatial Autocorrelation and Bayesian Spatial Statistical Method for Analyzing Intersections Prone to Injury Crashes

Identification of hazardous road locations (i.e., crash hot spots or black spots) is a standard practice in departments of transportation (DOTs) throughout the United States. Often, DOTs use relatively straightforward methods to detect high crash sites; these methods, however, may not be accurate. The objectives of this study are to develop two different methodologies: (a) a geographic information system (GIS)-based method to detect crash hot spots and (b) a spatial regression method with structured and unstructured random effects in continuous space to investigate the intersection-level factors that influence the concentration of fatal injury crashes. Using injury and property damage only (PDO) crash data from Tucson, Arizona, this study shows that spatial dependence plays a strong role during the analyses of road-traffic crashes. These spatial dependencies, accounted through spatial autocorrelation, help to detect statistically significant clusters of crashes involving fatalities or severe or minor injuries, in a GIS framework. This study also develops statistical models of severe- and minor-injury crashes, using classical negative binomial (NB) and Bayesian spatial statistical methods incorporating spatially structured and unstructured random effects. The coefficient estimates from a Bayesian framework are similar to those from NB estimation, but with better precision. The model that includes spatial correlations also indicates the potential to reduce the bias associated with model mis-specification by changing the estimate of the annual average daily traffic coefficients. This studys results also indicate that spatially structured correlation is quite significant in cases of crashes involving minor injury or PDO and that unstructured effects are somewhat significant at the intersection-level for cases of severe-injury crashes.

Study of intersection accidents by maneuver type

Studies dealing with the effect of road geometry on accidents by vehicle maneuvers have been reported, mostly for western countries and a few for Asia. However, no such studies have been reported for Singapore. Traffic accidents arising from head-to-side and head-to-rear maneuvers at four-legged signalized intersections in Singapore were investigated. Based on accident data at intersections in the southwestern part of Singapore from 1992 to 1999, the factors affecting such accidents were explored using zero-altered probability models. Specific roadway geometries as well as traffic control and regulatory factors that influence the two categories of accidents were identified. It was found that head-to-side accidents tend to decrease if there is an adjacent intersection within 200 m and if bus stops along the approach are provided with bays. On the other hand, longer sight distances and the presence of a pedestrian refuge tend to increase this type of accident. Higher speed limits were found to reduce the instances of zero head-to-side accidents. It was also found that head-to-rear accidents decrease when the intersections are under adaptive signal control but increase when surveillance cameras are present. There is also some evidence to suggest that the presence of an uncontrolled left-turn channel, the existence of medians wider than 2 m, higher approach volumes, and more phases per cycle all contribute to higher instances of accidents by both maneuver types.

Fuzzy Cluster–Based Method of Hotspot Detection with Limited Information

In the absence of geometric design and traffic data, hot-spot identification (HSID) is done primarily with crash data only, based on techniques such as crash frequency (CF), fatal crash frequency (FCF), or equivalent property damage only (EPDO), despite the known limitations of these techniques. In this article, the authors propose an improved HSID technique that may be used with crash data only. Using disaggregate crash history information, this method estimates probabilities of crash severities by the major contributing factors using severity models. These probabilities are used to compute expected numbers of severe and fatal crashes at various locations which are then used to classify the locations into two fuzzy cluster, namely hotspot and non-hotspot using Fuzzy C-Means (FCM) algorithm. The identified hotspots are ranked based on their mean departure from the core of the hotspot cluster. These rankings are compared with rankings done using existing techniques namely CF, FCF, EPDO, and Empirical Bayes’ (EB). The proposed method is found to be a robust method for hotspot detection with performance better than existing methods that use crash data only and comparable to the EB method.

Valuing Factors Influencing Bicycle Route Choice Using a Stated-Preference Survey

AbstractThis paper demonstrates a willingness-to-pay (WTP)–based approach to quantify a bicyclist’s perception toward a few key attributes related to bicycle route choice. To check for city-specifi...

Segmenting Highway Network Based on Speed Profiles

GPS Data from vehicles making trips on the highway are a valuable source of information for highway data analytics. In this article we propose an algorithm for segmenting the highway network into homogenous stretches in terms of vehicle speed profiles. We have GPS data of trucks plying across India, transmitted at an interval of 10 minutes, for thousands of trips. We identify break-points for individual trips and then cluster those break-points to obtain highway segment ends. We calculate the average velocity of vehicles traversing the regions between these segment ends, i.e. the highway segments. Then we merge the segments using an iterative minimum difference merging algorithm. The segments obtained thus are meaningful and may be utilized in optimal trip planning, infrastructure management and other decision making tasks.

An Approach to Tackle Urban Congestion and Vehicle Emission by Manipulating Transport Operations and Vehicle Mix

Emission from motorized vehicles is a major source of air pollution in urban areas. However, it varies significantly with vehicle technology, type of fuel used, operating conditions, vehicle mix, etc. Understanding the relationship amongst congestion levels in terms of Level of Service (LOS) policies, emission levels and traffic compositions is important for effective policy development for pollution reduction. This study adopted an integrated optimization model to understand this complex relationship with the help of suitable performance indices considering total emissions, fuel consumption, vehicle delays as well as capacity utilization of an intersection in Kolkata, India. SYNCHRO, a transportation operational analysis program is used to develop all the possible LOS thresholds. Twelve different traffic compositions are considered by modifying the share of vehicle categories. Emission inventories are generated using MOBILE, SYNCHRO and CRRI methods of emission calculation. To validate the emission inventories developed from these models, concentrations of the two major pollutants, Suspended Particulate Matter (SPM) and Oxides of Nitrogen (NOx) are collected from the intersection site using High Volume Air Sampler. Estimation of emission for base case by MOBILE yielded closest results with that of actual emissions estimated by the High Volume Air Sampler at the site. While comparing the performance indices, for the Kolkata intersection, LOS B is found to be the most effective operating point for combined emissions, fuel consumption and traffic congestion (delay at the intersection) point of view. There is also evidence that reduction in emission is associated with decreased share of motorcycles.

Effects of Access, Geometric Design, and Heterogeneous Traffic on Safety Performance of Divided Multilane Highways in India

ABSTRACT Safety of the national highways (NHs) has been a major concern in India. Using 5 years of crash data from a 65-km stretch of a divided multilane NH, this study employs random parameter panel data models to identify factors affecting total crashes, rear-end and head-on crashes. Besides geometric design elements, this study focuses on the effect of access management strategies such as provision of service lane and presence of median opening on crash types. To capture the effect of mixed traffic on crash occurrence, vehicle composition is considered in addition to average daily traffic (ADT). Results suggest that the effects of segment length and ADT are generally fixed and consistent across crash types, but there exists variation in safety performance of horizontal alignment, access management strategies, and type of vehicles. Although the coefficient for proportion of motorized two wheelers is found to be random for rear-end crashes, higher proportions of truck traffic are found to be always associated with higher head-on collisions. Although segments with service lanes are associated with fewer rear-end and head-on crashes, median opening presence is found to increase both of these crash types even though the effect on total crashes is random.

Identifying and Characterizing Truck Stops from GPS Data

Information about truck stops in highways is essential for trip planning, monitoring and other applications. GPS data of truck movement can be very useful to extract information that helps us understand our highway network better. In this paper, we present a method to identify truck stops on highways from GPS data, and subsequently characterize the truck stops into clusters that reflects their functionality. In the procedure, we extract the truck stoppage locations from the GPS data and cluster the stoppage points of multiple trips to obtain truck stops. We construct arrival time distribution and duration distribution to identify the functional nature of the stops. Subsequently, we cluster the truck stops using the above two distributions as attributes. The resultant clusters are found to be representative of different types of truck stops. The characterized truck stoppages can be useful for dynamic trip planning, behavior modeling of drivers and traffic incident detection.

Spatial decision tree for accident data analysis

Accident data analysis deals with identifying a set of conditions of accident occurrences and the importance of the corresponding implication. It is of prime importance because it gives an insight into the reasons behind the number of fatal and other major injuries. Accident data has an inherent spatial context associated with it, as the location of the accident has an important role to play in its severity. This paper aims at categorizing and analyzing the accident data and drawing some meaningful inferences, that are implicit to the data. A spatial decision tree based approach has been used and implemented to draw some useful conclusions, which are spatially relevant with the severity of the accident. The experimentation has been carried out on the accident dataset, collected from the National Highway (NH6) connecting Kharagpur and Kolkata, India. The results exhibit some latent patterns, useful for further accident management.