Implicit Scenario Mixture Models for Travel Time Estimation
ke wan, Princeton UniversityShow Abstract
Alain Kornhauser, Princeton University
Dependent structure between the travel times between
different links are modeled via Gaussian
A Methodology to Back-calculate Individual Speed Data Originally Aggregated by Road Detectors
Lorenzo Catani, Politecnico di TorinoShow Abstract
Marco Bassani, Politecnico di Torino
Cinzia Cirillo, University of Maryland
Highway and traffic engineers collect vehicular speed data to support analysis and design activities, using detectors based on a variety of fixed and mobile device technologies. Most of the acquisition units aggregate speed data into speed classes for ease of management and storage. Unfortunately, this practice entails a significant loss of content associated with individual data. Moreover, individual speeds are often necessary in support of road safety analysis and speed management decisions.
To bridge this gap, the paper introduces an algorithm that disaggregates speed data collected using automatic road detectors which can only measure speed frequency in intervals. The objective is to obtain back‑calculated individual speeds to operate with continuous distribution functions rather than discrete ones. This allows the derivation of more robust, basic descriptive measures (average, variance, and percentiles) according to Normal, LogNormal, and Gamma probability distribution functions. Therefore, the information produced is more useful than that calculated from standard aggregated speed reports.
In this investigation, individual speed data collected from video cameras were used to derive reference distributions and descriptive measures on the same road sections where inductive double-loop detectors were installed. Comparisons between the back-calculated individual speeds and those collected from video cameras support the validity of the proposed algorithm.
Where Can Conflicts Be Surrogates for Crashes? An Investigation Based on a Semi-Parametric Statistical Approach
Yaohua Zhang, University of ConnecticutShow Abstract
Nalini Ravishanker, University of Connecticut
John Ivan, University of Connecticut
Sha Mamun, University of Connecticut
There is growing interest among traffic engineers for using so-called surrogate measures of safetyas an alternative to crash counts, especially in contexts where the temporal and/or spatial crash occurrence rate is extremely low. In order to facilitate this paradigm shift, it is useful to demonstrate significant association between conflicts and crashes, and to study how this association might vary by location. We investigate a semiparametric statistical approach called the tau-path method that enables us to rank locations by decreasing magnitude of the association between crash and conflict counts. We demonstrate the method in the context of pedestrian safety at intersections in central Connecticut with variation in several characteristics including crossing distance, pedestrian signal phasing type, presence or absence of on-street parking and surrounding land use. Locations with high association between conflicts and crashes were more likely to have exclusive pedestrian phasing and on-street parking. Among these locations, those with high conflict and crash counts were more likely to have on-street parking and be in non-residential areas. The tau-path method can be easily applied to other road safety contexts beyond investigating conflict counts as surrogates for crash counts. This approach is also relevant to more general data mining settings where there is a need to identify a subpopulation in which there is a strong association between a pair of variables of interest.
New Mixed MNP Model Accommodating a Variety of Dependent Nonnormal Coefficient Distributions
Chandra Bhat, University of Texas, AustinShow Abstract
Patricia Lavieri, University of Melbourne
In this paper, we propose a general copula approach to accommodate non-normal continuous mixing distributions in multinomial probit (MNP) models. In particular, we specify a multivariate mixing distribution that allows different marginal continuous parametric distributions for different coefficients. The effectiveness of our formulation is demonstrated through a simulation exercise and an empirical mode choice application in which we test different distributions for the travel cost and travel time coefficients.
Crash Count Modeling and Joint Injury-Severity Analysis Using Nonparametric Bayesian Models
Shahram Heydari, Imperial College LondonShow Abstract
Liping Fu, University of Waterloo
Lawrence Joseph, McGill University
Luis Miranda-Moreno, McGill University
In transportation safety studies, it is often necessary to account for unobserved heterogeneity and multimodality in data. The commonly used standard or over-dispersed generalized linear models (e.g., negative binomial models) do not fully address unobserved heterogeneity, assuming that crash frequencies follow unimodal exponential families of distributions. This paper employs Bayesian nonparametric Dirichlet process mixture models demonstrating some of their major advantages in transportation safety studies. We examine the performance of the proposed approach using two case studies. We compare the proposed model with other models commonly used in road safety literature including the Poisson-Gamma, random effects, and conventional latent class models. We use pseudo Bayes factors as the goodness-of-fit measure. In a multivariate setting, we extend the standard multivariate Poisson-lognormal model to a more flexible Dirichlet process mixture multivariate model. We allow for interdependence between outcomes through a nonparametric random effects density. Finally, we demonstrate how the robustness to parametric distributional assumptions (usually the multivariate normal density) can be examined using a mixture of points model when different (multivariate) outcomes are modeled jointly.
Stochastic Cusp Catastrophe Models with Traffic and Weather Data for Crash Severity Analysis on Urban Arterials
Athanasios Theofilatos, Loughborough UniversityShow Abstract
George Yannis, National Technical University of Athens (NTUA)
Eleni Vlahogianni, National Technical University of Athens (NTUA)
The investigation of crash severity with freeway traffic and weather data has recently received significant attention by researchers. This paper extends previous research by proposing nonlinear models for modeling crash injury severity enhanced with traffic and weather data collected from urban arterials in Athens, Greece. Cusp catastrophe models are applied and compared with traditional statistical models. The results of crash severity models support the potential applicability of the cusp catastrophe theory to road safety, at least when crash severity is expressed as the number of severely and fatally injured by total number of persons involved in a crash. Variations in speed, average flow upstream of the location of interest, crash type and wind speed, were found to have a potential effect on the system dynamics. However, findings do not always confirm the strong presence of nonlinearity. When crash severity is expressed as the number of injured persons by the total number of vehicles involved in a crash, linear models could also be used to describe the underlying phenomenon.
Modeling Multinomial Outcomes from Partner Selection and Joint Decision-Making Processes
Dapeng Zhang, Virgin Hyperloop OneShow Abstract
Xiaokun (Cara) Wang, Rensselaer Polytechnic Institute (RPI)
Burgeoning information technology innovations and wide adoptions of global positioning system (GPS) devices have greatly changed the transportation system. For travelers, the real-time ridesharing platforms allow drivers and riders to interact, pair up, and jointly decide on departure time and routes. In freight transportation, the prosperity of e-commerce leads to individualized real-time seller-buyer matching and their joint decisions on delivery modes and time windows. Transportation agents mutually select, or get matched with their counter-partners, and jointly make decisions on a set of matters. Existing econometric models are not able to behavioral-consistently capture these new phenomena which include intricate matching networks, mutual selection, and intensive joint decision making. This paper develops an innovative econometric model to fill the void. The proposed model consists of two parts: The first part explains the matching process in a many-to-one matching structure; the second part characterizes the joint decision making process of mutually-selected decision makers. The two parts are integrated by recognizing their dependency by a sample selection process: a joint response is only observed for matched decision makers. The proposed model is estimated using a Bayesian Markov-Chain Monte-Carlo approach with data augmentation. The likelihood functions and posterior distributions are derived, followed by a set of simulation studies to test parameter recovery capability at different parameter settings.
Negative Binomial Crash Sum Model for Time-Invariant Heterogeneity in Panel Crash Data: Some Insights
Ghasak Mothafer, Nagoya UniversityShow Abstract
Toshiyuki Yamamoto, Nagoya University
Venkataraman Shankar, Texas Tech University
This paper presents a negative binomial crash sum model as an alternative for modeling time invariant heterogeneity in short crash data panels. Time invariant heterogeneity arising through multiple years of observation (2005-2007) for each segment is viewed as a common unobserved effect at the segment level, and typically treated with panel models involving fixed or random effects. Random effects model unobserved heterogeneity through the error term, typically following a gamma or normal distribution. We take advantage of the fact that gamma heterogeneity in a multi-period Poisson count modeling framework is equivalent to a negative binomial distribution for a dependent variable which is the summation of crashes across years. The Poisson panel model referred to in this paper is the random effects Poisson gamma (REPG). In the REPG model, the dependent variable is an annual count of crashes of a specific type. The multi-year crash sum model is a negative binomial (NB) model that is based on three consecutive years of crash data (2005-2007). In the multi-year crash sum model, the dependent variable is the sum of crashes of a specific type for the three-year period. Four categories (in addition to total crashes) of crash types are considered in this study including rear end; sideswipe; fixed objects and all-other types. The empirical results show that, the three-year crash sum model is a computationally simpler alternative to a panel model for modeling time invariant heterogeneity, while imposing fewer data requirements such as annual measurements.
Truncated Bayesian Nonparametric Modeling of Multistate Travel Time Distribution
Emmanuel Kidando, Florida State UniversityShow Abstract
Ren Moses, Florida State University
Eren Ozguven, FAMU-FSU College of Engineering
Thobias Sando, University of North Florida
Multi-state models are preferred over single-state probability models in modeling the distribution of travel time. Literature review indicated that the finite multi-state modeling of travel time using lognormal distribution was superior to other probability functions. In this study, we extend the finite multi-state lognormal model in estimating the travel time distribution to unbounded lognormal distribution. In particular, a non-parametric Dirichlet Process Mixture Model (DPMM) with stick-breaking process representation was used. The strength of the DPMM is that it can choose the number of components dynamically as part of the algorithm during parameter estimation. To reduce computational complexity, the modeling process was limited to a maximum of six components. Then, the Markov Chain Monte Carlo (MCMC) sampling techniques were employed to estimate the posterior distribution of the model parameters. Speed data from nine links of a freeway corridor, aggregated on 5-minutes basis, were used to calculate the travel time on each link. The results demonstrated that this model offers significant flexibility in modeling to account for complex mixture distributions such as travel time without specifying the number of components. The DPMM modeling further revealed that freeway travel time is characterized by multi-state and single-state depending on the inclusion of onset and offset of congestion periods. The Kolmogorov-Smirnov hypothesis test of the model was conducted and the results showed a reasonable fit.
Joint Framework for Static and Real-Time Crash Risk Analysis
Shamsunnahar Yasmin, Centre for Accident Research & Road Safety – QueenslandShow Abstract
Naveen Eluru, University of Central Florida
Ling Wang, Tongji University
Mohamed Abdel-Aty, University of Central Florida
The current research effort bridges the gap between traditional crash risk and real-time crash risk models by developing a joint model that accommodates for both dimensions in developing crash risk analysis models. Specifically, we develop a joint reactive and proactive crash modeling framework by coupling the static monthly crash risk and dynamic real-time crash risk in a unified econometric framework for a microscopic analysis unit. In the joint modeling approach, we propose and estimate an alternative to the case-control binary logit based real-time crash risk analysis by proposing a multinomial logit based approach where time periods serve as alternatives and the chosen alternative is the time period in which crash occurs. The joint model also allows us to accommodate for the common unobserved factors that increase the likelihood of a crash in microscopic unit to affect the real-time crash risk propensity. We demonstrate the application of the proposed approach by using data on roadway segments from three expressways in Central Florida (State Roads 408, 417, and 528) for 29 months. The monthly crash risk component is examined by using binary logit model employing different static roadway attributes (roadway geometry and operational attributes). While the real-time crash risk component is examined by using a random utility model employing different dynamic traffic attributes (volume, speed, lane occupancy and environmental conditions). The outcome of the proposed approach allows us to predict both the static and dynamic crash risk components simultaneously in a single econometric framework.
An Introduction to the Network Weight Matrix
Alireza Ermagun, Northwestern UniversityShow Abstract
David Levinson, University of Sydney
This study introduces the network weight matrix as a replacement for the spatial weight matrix to measure the spatial dependence between links of a network. This matrix stems from the concept of betweenness centrality and vulnerability in network science. The elements of the matrix are a function not simply of proximity, but of network topology, network structure, and demand configuration. The network weight matrix has distinctive characteristics, which are capable of reflecting spatial dependence between traffic links: (1) The elements are allowed to have negative and positive values, which capture competitive and complementary nature of links, (2) the diagonal elements are not fixed to zero, which takes the self-dependence of a link upon itself into consideration, and (3) the elements not only reflect the spatial dependence based on the network structure, but they acknowledge the demand configuration as well. We verify the network weight matrix by modeling traffic flows in a 3 by 3 grid test network with 9 nodes and 24 directed links connecting 72 origin-destination (OD) pairs. The results disclose models encompassing the network weight matrix outperform both models without spatial components and models with the spatial weight matrix. This leads to the conclusion that the network weight matrix represents a more accurate and defensible spatial dependency between traffic links, and thereby augments traffic flow prediction.
Grouped Response Ordered Logit Count Model Framework
Naveen Eluru, University of Central FloridaShow Abstract
Shamsunnahar Yasmin, Centre for Accident Research & Road Safety – Queensland
The study proposes and estimates a new econometric
framework for analysing crash count events labeled as Grouped Response Ordered
Logit Count (GROLC) model. The proposed GROLC model builds on earlier work on
generalized ordered logit count (GOLC) model proposed recently. The proposed
framework relates the crash count propensity to the observed counts directly
while also accommodating for heteroscedasticity. The proposed model is
demonstrated by using Traffic Analysis Zonal (TAZ) level bicycle crash count
data for Montreal. The mode framework employs a comprehensive set of exogenous
variables − accessibility measures, exposure measures, built environment, road
network characteristics, sociodemographic and socioeconomic characteristics.
Further, we also compare the performance of the proposed model to the most
commonly used Negative Binomial (NB) model and the GOLC model by generating a
comprehensive set of measures to evaluate model performance and data fit across
these frameworks. The comparison exercise clearly highlights the superiority of
the GROLC model in terms of data fit compared to NB and GOLC model in current
study context. The fit measures for comparing the predictive performance also
indicate that GROLC model offers superior predictions both at the aggregate and
disaggregate levels. Overall, the results from this comparison exercise points
out that GRLOC offers improved performance in the context of crash count
modeling and thus is a promising alternate econometric framework.
Multimodal Crash Frequency Modeling: Multivariate, Multivariate Spatial, Multivariate Temporal, or Multivariate Spatial-Temporal?
Wen Cheng, California State Polytechnic University, PomonaShow Abstract
Gurdiljot Gill, California State Polytechnic University, Pomona
Roya Falahati, California State Polytechnic University, Pomona
Xudong Jia, California State Polytechnic University, Pomona
Jiao Zhou, 1978
Tom Vo, Southern California Association of Governments
A multimodal approach can help build more comprehensive safety portraits and ease the task of selecting sites for safety improvement interventions. A central issue to the successful implementation of multimodal approach is the development of appropriate crash frequency models which can jointly estimate the crash risk of different mode users. The study proposed two multivariate spatial-temporal models to analyze the modal crash data: one with fixed time trend applied to all modes, the other with mode-varying time trend coefficients. The proposed models were then compared with three types of multivariate models used in the past studies. The major objective is to examine the benefits of the newly proposed models which have substantially increased computational cost since both dimensions of time and space are considered, as well as their interactions. Moreover, the relative site ranking performance among the alternative models were also evaluated with different evaluation criteria.
Seven years of crash data and other covariates associated with traffic analysis zones in the City of Irvine, California were collected for modeling assessment purpose. The modeling results indicated that proposed multivariate space-time models had the superior performance while the multivariate without accounting for spatial and temporal correlations performed the worst. The site ranking evaluation revealed the strong positive correlation between siting ranking and modeling performance. Overall, the study is anticipated to enhance the understanding of safety impacts from the interaction of various modes.
Detecting Changes in Accident Rates Using a Hierarchical Bayesian Approach: An Application to the I-710 and the implementation of the PierPASS Program
Ankoor Bhagat, University of California, IrvineShow Abstract
Jean-Daniel Saphores, University of California, Irvine
R. Jayakrishnan, University of California, Irvine
Road accidents involving heavy duty trucks have long been of concern but detecting the impact on accidents of various policies is typically challenging. The objective of this study is to understand if there was a change in accident rates on a busy freight corridor (the I-710 freeway in Los Angeles County, California) connecting the Ports of Los Angeles and Long Beach to nearby intermodal rail, trans-loading facilities, and warehouses after the implementation of the PierPASS program on July 23, 2005. We analyzed 2,043 accidents that occurred in 2005 on the I-710 freeway; approximately 27.8 percent of these accidents involved trucks. We estimated a three stage hierarchical Bayesian change point model with MCMC developed by Carlin et al. (1992) to evaluate whether the implementation of the PierPASS program resulted in a change of accident rates on the I-710. After successfully verifying and validating our model on the dataset used by Carlin et al. (1992), and Raftery and Akman (1986), we analyzed road accidents on the I-710 for 2005 filtered by four (approximately 6 mile) segments and four time periods corresponding to different traffic regimes. We generated the probability distribution of the difference between accident rate parameters and built 95% High Density Intervals. Results indicate that there was no significant change in accident rate in 2005 following the implementation of PierPASS. To our knowledge, this is the first time that a three stage hierarchical Bayesian change point model with MCMC was applied to a transportation problem.