Xin Ye

A simultaneous equations model of crash frequency by severity level for freeway sections

This paper presents a simultaneous equations model of crash frequencies by severity level for freeway sections using five-year crash severity frequency data for 275 multilane freeway segments in the State of Washington. Crash severity is a subject of much interest in the context of freeway safety due to higher speeds of travel on freeways and the desire of transportation professionals to implement measures that could potentially reduce crash severity on such facilities. This paper applies a joint Poisson regression model with multivariate normal heterogeneities using the method of Maximum Simulated Likelihood Estimation (MSLE). MSLE serves as a computationally viable alternative to the Bayesian approach that has been adopted in the literature for estimating multivariate simultaneous equations models of crash frequencies. The empirical results presented in this paper suggest the presence of statistically significant error correlations across crash frequencies by severity level. The significant error correlations point to the presence of common unobserved factors related to driver behavior and roadway, traffic and environmental characteristics that influence crash frequencies of different severity levels. It is found that the joint Poisson regression model can improve the efficiency of most model coefficient estimators by reducing their standard deviations. In addition, the empirical results show that observed factors generally do not have the same impact on crash frequencies at different levels of severity.

Departure-Time Choice and Mode Choice for Nonwork Trips: Alternative Formulations of Joint Model Systems

Modeling travel demand by time of day is gaining increasing attention in the practice of travel demand forecasting. The relationship between time-of-day (departure-time) choice and mode choice for nonwork trips is investigated. Two alternative causal structures are considered: one in which departure-time choice precedes mode choice and a second in which mode choice precedes departure-time choice. These two causal structures are analyzed in a recursive bivariate probit modeling framework that allows random error covariance. The estimation is performed separately for worker and nonworker samples drawn from the 1999 Southeast Florida Regional Household Travel Survey. For workers, model estimation results show that the causal structure in which departure-time choice precedes mode choice performs significantly better. For nonworkers, the reverse causal relationship, in which mode choice precedes departure-time choice, is found to be a more suitable joint modeling structure. These two findings can be reasonably explained from a travel behavior perspective and have important implications for advanced travel demand model development and application.

Joint Model of Vehicle Type Choice and Tour Length

Tour-based microsimulation model systems are increasingly being applied to the forecasting of travel demand. This paper examines the relationship between two dimensions of tours: the type of vehicle (in a household that owns multiple vehicles of different types) chosen to undertake the tour and the overall length (distance traveled) of the tour. These two dimensions are of much interest in the current planning context, in which concerns about energy sustainability and greenhouse gas emissions are motivating planners to seek ways to mitigate the adverse impacts of automotive travel. Moreover, virtually all tour-based models currently used do not explicitly account for choice of vehicle type in the modeling of tour attributes, despite the critical importance of the choice of vehicle type for energy and emissions analysis. This paper presents a joint discrete–continuous model of choice of vehicle type and length of tours. Estimation results suggested that significant common unobserved factors affected vehicle type choice and length of tours. These factors justified the use of modeling approaches with joint simultaneous equations to model tour attributes. The model specification in which vehicle type choice affected tour length performed better than the specification in which tour length affected vehicle type choice. This outcome suggested that choice of vehicle type (and allocation to household members) was a longer-term choice that influenced shorter-term tour-length choices.

A Probit-based Joint Discrete-continuous Model System: Analyzing the Relationship between Timing and Duration of Maintenance Activities

This paper presents a probit-based discrete-continuous modeling methodology to analyze relationships between discrete and continuous choice dimensions often encountered in activity-travel behaviour research. The probit-based approach allows one to adopt a flexible multivariate normally distributed error covariance structure that overcomes the limitations associated with other discrete-continuous modeling methods that rely on two-step limited information techniques or restrictive distributional transformations. The paper presents the detailed formulation of the modeling methodology and demonstrates its applicability through an analysis of the relationship between the timing (scheduling) and duration of household maintenance activities that include shopping and personal business activity episodes. A new non-nested test developed by the authors is used to compare alternative model structures and identify the nature of the joint relationship between the timing and duration of maintenance activities. Models are estimated separately for commuter and non-commuter samples drawn from the 2000 Switzerland Microcensus of Travel. Model estimation results show that error covariances are significant for commuter models of maintenance activity timing and duration. Non-nested model comparisons indicate that the model specification where time-of-day choice affects activity duration offers a statistically superior goodness-of-fit in comparison to the model specification where activity duration affects time-of-day choice. These findings lend credence to the notion that the joint relationship between timing and duration adopted in activity-based model systems should be one in which the activity schedule or agenda drives activity time allocation or duration.

Formulation of an Activity-Based Utility Measure of Time Use: Application to Understanding the Influence of Constraints

This paper presents and demonstrates a methodology to compute a composite time use utility measure that accounts for in-home and out-of-home activity engagement and time allocation patterns of individuals. The measure could be used for welfare analysis in the context of a policy intervention and to model the search and adaptation routine that individuals may follow in choosing an alternative activity travel pattern in response to a policy intervention. The proposed measure can be implemented as a postprocessor for activity-based model systems to evaluate the satisfaction that travelers derive from their overall daily activity travel pattern. With data from the 2005 American Time Use Survey, the analysis was performed for a sample of women stratified by employment status, income, and presence of children. Comparisons of time use utility measures across these cross-classified groups offer insights into the influence of temporal (employment), monetary (income), and household obligation (children) constraints on the utility individuals derive from their activity travel pattern. In general, it was found that time use utility values were affected most adversely by temporal constraints, followed by monetary constraints, and then by the presence of children.

An Examination of the Endogeneity of Speed Limits and Accident Counts in Crash Models

A properly set speed limit establishes a reasonable and acceptable threshold that the majority of drivers can follow. Much literature has been devoted to investigating the relationships between speed limit and accident number, but the results have been widely variable. It is speculated that the variance of these conclusions can be attributed to the endogeneity of speed limit and accident count. Traffic volumes and crash counts at a total of 298 intersections in the City of Corona were collected and analyzed using simultaneous equation models to eliminate the influence of the endogenous variables and obtain unbiased predictor variables. By running single-equation models individually involving crash counts and speed limits and then comparing them with a simultaneous equation model that evaluates these same variables, it was possible to determine the effect of endogeneity on the resultant estimator variables. It was found that although the difference between the estimator variables in the single and simultaneous equation models was not statistically significant in the locations observed in this study, the presence of endogenous variables was confirmed. It is therefore anticipated that endogeneity might need to be accounted for in transportation models involving crash histories and speed limits in the future.

Synthetic Environment to Evaluate Alternative Trip Distribution Models

In this paper an environment is synthesized to incorporate spatial distributions of population and employees and to simulate travelers’ destination choice behaviors following utility-maximization decision rules. In this synthetic environment, two alternative trip distribution models—the destination choice model and the gravity model—are evaluated by comparing estimated model coefficients and trip matrices against their true counterparts. The destination choice model provides reasonable model coefficients and trip matrix when the average trip length is much greater than the zone size. However, when the average trip length comes closer to the zone size as a result of significant spatial aggregation errors the model coefficients appear more biased, and more errors occur in trip matrix estimation. In the gravity model, linear regression does not provide consistent coefficients for trip attraction variables and therefore cannot accurately estimate trip attractions. It is not optional but necessary to apply the destination choice model for consistently estimating the trip attraction and trip matrix in trip distribution.

Accelerated Procedure of Multiclass Highway Traffic Assignment for Maryland Statewide Transportation Model

The Maryland Statewide Transportation Model (MSTM) under development contains 20 user classes in the highway assignment procedure, including three types of autos: single-occupancy vehicles (SOV), high-occupancy vehicles (HOV) with two passengers, and HOV with three or more passengers classified by five income levels: commercial vehicle, median trucks, heavy trucks, regional autos, and regional trucks. A comprehensive disaggregation of user classes aids in capturing differences in how the transportation network is used. For example, SOV drivers are not allowed to use HOV lanes; trucks cannot be driven on truck-prohibited lanes; drivers at various income levels make different decisions about using toll roads because of their different values of time. However, a large number of user classes impose substantial computational burden on the procedure of multiclass highway assignment. If the conventional algorithm is used, the time consumption for this procedure is approximately proportional to the total number of user classes. In this study, a novel algorithm proposed by Robert Dial in 2006 is used to transfer min-path trees from one user class to another in the procedure of multiclass traffic assignment. This algorithm made it possible to speed up the highway assignment procedure of MSTM by a factor of 5.87.

Investigation of Underlying Distributional Assumption in Nested Logit Model Using Copula-Based Simulation and Numerical Approximation

This paper proposes a practical approach for drawing random samples that satisfy the underlying distributional assumptions in the nested logit model. This approach is a combined application of a series of statistical and numerical techniques including copula-based characterization of joint distributions, quasi-Monte Carlo integration, coordinate conversion, dichotomous approximation, and linear interpolation. With this approach, one may efficiently sample from and directly observe the distribution of common random components that result in correlation between random utilities in the nested logit model. Monte Carlo studies are conducted for two- and three-level nested logit models to evaluate the performance of random samples to recover model coefficients in nested logit models. Simulation experiments show that the relative differences are no more than 3% between mean values of estimators and their true values, indicating the proposed approachs high level of accuracy in reproducing random components in nested logit models.