2021

Jason Hawkins, University of Toronto
Khandker Nurul Habib, University of Toronto

In this paper, we provide a comparison of implementations of Bayesian estimation of mixed multinomial logit (MMNL) models. Our objective is to provide a systematic comparison of the runtime, efficiency, and model implementation details associated with several alternative workflows. Analysis is based on three case studies. We argue that previous comparisons in the transportation literature have lacked appropriate metrics for comparison. Effective sample size statistics are proposed as a means of accurately comparing the ability of Bayesian samplers to generate independent draws for use in statistical inference. The Allenby-Train algorithm implemented in the R package Apollo is compared with the NUTS sampler implemented in Stan. While the Allenby-Train algorithm tends to generate draws much faster than NUTS, we find that the high correlation between success samples makes the two methods comparable. In addition to traditional MCMC sampling, we also examine the method of variational Bayes (VB) in terms of speed and ability to reproduce a known set of parameters. Our assessment is that the low speed of NUTS has been overstated in the transportation literature. Finally, the ability of probabilistic programming languages, such as Stan, to facilitate rapid development of a range of models (beyond MMNL) and perform diagnostics offers important advantages.

Paras Agrawal (agrawal.paras94@gmail.com), Asian Institute of Technology
Surachet Pravinvongvuth, Asian Institute of Technology

Hyperloop is being looked at as the future of high-speed transportation, however there is not much of published information available for the demand estimation for Hyperloop. Hyperloop is seen as an ideal mode of transport which could be both fast and efficient unlike any other mode of transport, therefore, the principal objective of this paper is to present the estimation of the travel demand for Hyperloop for Bangkok- Chiang Mai sector in Thailand using traveler surveys. Stated Preference (SP) survey was conducted for Air and Car travelers along the Bangkok- Chiang Mai sector and a Nested Logit (NL) model is developed to understand the factors affecting the mode choice and to predict the number of travelers willing to shift to Hyperloop from Air and Car. The model is further utilized to perform elasticity analysis, to determine value of travel time savings and to obtain ridership estimates with Hyperloop scenario. The highest value of elasticity for Hyperloop is obtained for total travel cost followed by total travel time and monthly income of the travelers. It is found that the passengers who prefer to travel by Hyperloop are willing to spend 515.54 THB to save a unit hour. Further, the number of trips per year by Hyperloop for 2020 and 2025 are estimated at 49.34% and 51.41% of the total trips respectively. It was observed that from the day one of operations of Hyperloop, it will be the predominant mode of transportation between Bangkok-Chiang Mai sector followed by Air and Car.

Lennart Oelschläger, Bielefeld University
Dietmar Bauer, Bielefeld University

Discrete choice models lie at the heart of many transportation models. This holds true for mode choice models as well as for models of vehicle purchase decisions, to name just a few applications. The standard random utility models often fail to take into account heterogeneity of individual deciders, compare for example Train (1), Chapter 6, or Train (2). Individual heterogeneity in preferences for different deciders has been modelled by imposing mixing distributions on the coefficients. However, the literature does not provide much guidance so far for the specification of the mixing distributions apart from trial and error procedures using several standard parametric distributions. Based on the ideas of Train (2) and Bhat and Lavieri (3) recently Bauer et al. (4) introduced procedures for non-parametrically estimating mixed multinomial probit models. While these procedures have been demonstrated to be useful in cross-sectional context, the evaluation of the probit likelihood in the panel setting proofs numerically challenging.  Extending Scaccia and Marcucci (5) we propose an approach that on the one hand is capable of approximating the underlying mixing distribution while on the other hand being numerically favorable since no likelihood function has to be evaluated. The approach uses a Bayesian framework for estimating a latent class mixed multinomial probit model where the number of latent classes is updated within the algorithm on a weight-based strategy. Presenting simulation results, we demonstrate that the approach is suitable for guiding the specification of mixing distributions in empirical applications.

Mehrdad Tajalli, North Carolina State University
Amir Mirheli, North Carolina State University
Ali Hajbabaie (ahajbab@ncsu.edu), North Carolina State University
Leila Hajibabai, North Carolina State University

This paper presents a set of predictive models to estimate the annual utilization of seven non-stationary highway equipment types based on a number of explanatory variables including their annual fuel cost, downtime hours, age, and weight. Furthermore, another set of models are fitted to predict the annual operational cost for these equipment types based on the most important contributing factors. The prediction models are developed after a nationwide data collection that was supported by the National Cooperative Highway Research Program (NCHRP) under project 13-05: Guide for Utilization Measurement and Management of Fleet Equipment. Several years of collected data from seven state Departments Of Transportation (DOT) are processed and used for model development. This research has identified the annual mileage as an appropriate utilization measurement metric that is widely used by many state DOTs. Various model structures to predict the annual mileage are considered. The logarithmic function of annual mileage has provided the most appropriate structure. The final annual mileage predictive models have R-squared values that are between 0.65-0.89, which indicates a good fit for all models. The models are validated by performing several statistical tests and they have satisfied all required assumptions of regression analysis.

Roland Varriale, Argonne National Laboratory
Kenley Pelzer, Argonne National Laboratory
Michael Jaynes, Argonne National Laboratory
Eric Rask, Argonne National Laboratory

In order to achieve control over certain vehicle functionality, adversarial agents frequently must spoof the speed of a moving vehicle. One way to combat this is to poll numerous signals from the vehicle network in order to evaluate the validity of a given signal. This research used discrete correlation function calculations to find pairs of signals within an automotive CAN bus network that exhibit strong correlation during normal driving conditions. This project uses data collected from a vehicle CAN bus and calculates the discrete correlation function values at various lag inputs, and analyzes how this information might be used to better protect against cyber security threats against moving vehicles.

Dawei Li, Southeast University
Siqi Feng (213171238@seu.edu.cn), Southeast University

Recently, some logit-type, closed-form asymmetric choice models have been proposed by literature for solving the class imbalance problem. However, these models have rarely been applied to route choice modeling by real world data. In this paper, we propose three specification approaches for the asymmetric parameters in these models. Then, we estimate four asymmetric choice models by GPS data collected in China (taxi) as well as Israel (bicycle), and illustrate the behavioral distinctions in different route choice contexts. The results confirmed the existence of class imbalance in route choice, as the multinomial scobit and uneven logit models perform well and stably with different parameter specifications, while the other two asymmetric models also have acceptable performances with respect to the path-size logit (PSL) model. Meanwhile, the performances of the specification approaches are case dependent, among which the combined method often outperforms the other two and is thus recommended. Note that some parameter specifications may cause changes in the estimates' signs and weaken the model's interpretability.

Vishal Mahajan (vishal.mahajan@tum.de), Technische Universitat Munchen
Emmanouil Barmpounakis, Ecole Polytechnique Federale de Lausanne (EPFL)
Md Alam, Technische Universitat Munchen
Nikolas Geroliminis, Ecole Polytechnique Federale de Lausanne (EPFL)
Constantinos Antoniou, Technische Universitat Munchen

Advancements in hardware and software technologies have given rise to novel and innovative methods of data collection, also for traffic management and road safety. Use of Unmanned Aerial Systems is a relatively new method for collecting traffic data. Data captured from drones is expected to have potential applications due to its unique advantages such as increased observability over an area and recording of naturalistic driving behavior compared to traditional methods. Data filtering or specifically anomaly detection is an important step during the data processing so that the data is fit for use. To the best of our knowledge, anomaly detection from a large drone dataset and that too from an urban area has not been tried before. In this study, we use pNEUMA dataset, which is a large dataset with almost 0.5 million trajectories captured by a swarm of drones over congested streets of downtown Athens, Greece. This dataset is novel and offers a fresh opportunity to demonstrate noise and outlier detection while also focusing on issues specific to the drones and urban driving context. We adapt an existing anomaly detection algorithm for our case and carry out a sensitivity analysis to fine-tune the algorithm. The results show that with this approach, several anomalous segments of the trajectory can be identified. The approach used in this study can help in treating the anomalous segments selectively instead of applying manual thresholds and global smoothing, which in some cases could lead to loss of information.

Anupriya - (anupriya15@imperial.ac.uk), Imperial College London
Daniel Graham, Imperial College London
Prateek Bansal, Imperial College London
Daniel Hörcher, Imperial College London
Richard Anderson, Imperial College London

During peak hours, metro systems often operate at very high service frequencies to transport large volumes of passengers. However, the punctuality of such operations is severely impacted by a vicious circle of passenger-congestion and train delays. In particular, high volumes of passenger boardings and alightings may lead to increased dwell times of trains at various stations, that may eventually cause queuing of trains in upstream. Such stations act as active bottlenecks in the metro network and congestion propagates from these bottlenecks to the entire network. Thus, understanding the mechanism that drives passenger congestion at these bottleneck stations is crucial to develop informed control strategies (for instance, ramp metering of passengers entering such stations) and operationalise them to prevent arising of delays. To this end, we conduct the first station-level analysis and estimate a causal relationship between train flow and passenger movement using smartcard and train movement data of Mass Transit Railway, Hong Kong. We adopt a fully-flexible non-parametric spline-based regression approach and apply instrumental variable based estimation to control for any confounding bias that may occur due to unobserved characteristics of metro operations. Through the results of the empirical study, we identify the bottleneck station and provide estimates of optimum passenger movements and service frequencies at such stations. These estimates, along with real data on daily demand, would help in devising the above-discussed control strategies.

Rafael Delpiano, Universidad de los Andes, Chile

There is growing interest in understanding the lateral dimension of traffic. This trend has been motivated by the detection of phenomena unexplained by traditional models and the emergence of new technologies. Previous attempts to address this dimension have focused on lane changing and non-lane-based traffic. The literature on vehicles keeping their lanes has generally been limited to simple statistics on vehicle position while models assume vehicles stay perfectly centered. Previously the author developed a two-dimensional traffic model aimed to capture such behavior qualitatively. Still pending is a deeper, more accurate comprehension and modelling of the relationships between variables in both axes. The present article is based on the NGSIM datasets. It was found that lateral position is highly dependent on the longitudinal position, a phenomenon consistent with data capture from multiple cameras. A methodology is proposed to alleviate this problem. Also discovered was that the standard deviation of lateral velocity grows with longitudinal velocity and that the average lateral position varies with longitudinal velocity by up to 8 cm, possibly reflecting greater caution in overtaking. Random walk models were proposed and calibrated to reproduce some of the characteristics measured. It was determined that drivers' response is much more sensitive to the lateral velocity than to position. These results provide a basis for further advances in understanding the lateral dimension. It is hoped that such comprehension will facilitate the design of autonomous vehicle (AV) algorithms that are friendlier to both passengers and the occupants of surrounding vehicles.

John Kodi, Florida International University
Emmanuel Kidando, Mercer University
Thobias Sando, University of North Florida
Priyanka Alluri, Florida International University

The adaptive signal control technology (ASCT) is a Transportation Systems Management and Operations (TSM&O) strategy that automatically and dynamically adjusts the signal timing parameters to optimize corridor performance based on real-time traffic demand. This study quantifies the mobility benefits of the ASCT using a Bayesian switch-point regression (BSR) model. The analysis was based on a 3.3-mile corridor along Mayport Road in Jacksonville, Florida. The results revealed that ASCT increases travel speeds by 4% on mid-week days (Tuesday, Wednesday, and Thursday) in the northbound direction. However, mixed results were observed in the southbound direction, which may be attributed to congestion and higher driveway density. The BSR model results revealed that there is a significant difference in the operating characteristics between the with and without ASCT scenarios. These findings may provide researchers and practitioners with an effective means for conducting an economic appraisal of the ASCT strategy, a key consideration for transportation agencies when planning future ASCT deployments.

Irfan Ahmed, University of Wyoming
Mohamed Ahmed, University of Wyoming

Rural highways fatality rates are always higher than urban highways ones. There are abundant studies in the literature explored the impact of weather effects and the consequent condition such as reduced visibility, wet pavement on crash injury severity. However, research on crash injury severity using disaggregate analysis based on weather conditions on rural two-lane two-way highways is limited. Such analysis provides useful insights to transportation planners in allocating resources based on weather conditions. In this study separate models for adverse and non-adverse weather conditions were developed using crash data of rural two-lane two-way highway corridor in two neighboring states, Wyoming and Colorado. The years of crash data used in this study were between 2007 and 2016. A random-intercept Bayesian logit approach was used to analyze the dichotomous injury severity response and capture the heterogeneity in the variance of the random parameter. An efficient Markov Chain Monte Carlo sampling technique known as No-U-Turn Hamiltonian Monte Carlo was employed to sample the posterior distributions of the parameter estimates. Likelihood Ratio tests indicated that the use of separate models was justified with over 95% confidence. The model results provided valuable insights into the contributing factors of crashes occurring in adverse and non-adverse weather conditions. Findings from the separate models suggest that there are major differences in both the combination and magnitude of the significant contributing factors. The findings and the recommendations from this study could potentially be used to help guide the respective agencies in formulating injury severity mitigation policies and strategies.

Ali Etezady (etezady@gatech.edu), Georgia Institute of Technology (Georgia Tech)
Patricia Mokhtarian, Georgia Institute of Technology (Georgia Tech)
Giovanni Circella, University of California, Davis

This study investigates the latent patterns in the modal impacts of ridehailing services in a sample of Californian ridehailers, and how shared ridehailing usage frequency is associated with these ridehailing modal impact patterns. Using a dataset collected in Fall 2018, we use a latent class with distal outcome approach to firstly identify the latent classes of ridehailing modal impacts, and then analyze the relationship between the identified latent classes and shared ridehailing usage while controlling for other factors that directly influence the usage frequency. Our analysis points to three latent classes of ridehailing modal impacts. In our first class, where ridehailers are younger, lower income, and more urbanite, a majority/plurality report a decline in the usage of taxicabs and transit services. In our second and third classes, where ridehailers are relatively older and higher income, a majority of ridehailers report no change in their use of other modes, with the difference that Class 3 ridehailers also report being users of transit (as opposed to Class 2, who are not), but ridehailing usage does not impact their transit usage. We further analyze the association between these latent classes and shared ridehailing usage, and find that 50% of the frequent shared ridehailing users (weekly users) in our sample are associated with Class 1 of the ridehailing modal impact. This analysis helps provide a more detailed picture of how ridehailing interacts with other transportation modes in different population segments, and further investigates the sustainability promise of shared ridehailing.

Dada Zhang, Northern Arizona University
Chun-Hsing Ho, Northern Arizona University
Fangfang Zhang, Northern Arizona University

The paper presents an approach using regression analysis and probability distribution fitting to analyze pavement sensing patterns and signals collected on the I-10 corridors in Phoenix, Arizona. A vehicle is equipped with four sensors placed on the top of the control arms of the vehicle and one sensor is inside of the vehicle to gather the data for analysis. The result of the multiple regression analysis shows that the mean of sensors differ in a logarithmic scale at significance level 0.05, which suggests that all sensors should be included for pavement condition assessments. The distribution models are fitted using the acceleration vibration and can be used to determine the threshold values by computing a specified percentile, for example 99th percent. The determination of thresholds varies based on the statistical analysis and the data falls in the remaining percent would indicate the pavement deterioration, which is called significant points in the paper. The ANOVA results show that there is an association between two variables (pavement temperatures and the number of significant points) at the significance level 0.05 which indicates the pavement temperature does play an important role in controlling pavement condition. Based on the Time-Series analysis and prediction, the pavements will be deteriorated if the maintenance and rehabilitation will not be scheduled. The paper concludes that using multiple regression analysis and distribution fitting method provides a promoting approach that can be used to help determine the level of different pavement conditions as well as predicting future performance.

Binya Zhang, University of Maryland, College Park

Traffic incidents on freeways cause a considerable loss of life and property. Some traffic operation organizations provide freeway safety services to improve the roadway’s safety condition by assisting in the detection and clearance of incidents. To offer assistance in time, the traffic operation centers generally use patrol vehicles to cover the freeway networks. However, the risk of an incident occurring may differ considerably among road segments in the networks. By giving these high-risk segments more resources, the efficiency of the freeway safety service may increase. Therefore, it is essential to recognize the road segments having higher incident risks among the whole freeway network. This research aims at providing a method of detecting the road segments with a higher risk of traffic incidents. The risk will be considered from both spatial and temporal factors by implementing the multi-kernel density estimation method. The statistics of spatial analysis will be applied to evaluate the detection results.

Mankirat Singh, California State Polytechnic University, Pomona
Wen Cheng, California State Polytechnic University, Pomona
Ranjithsudarshan Gopalakrishnan, California State Polytechnic University, Pomona
Bengang Li, California State Polytechnic University, Pomona
Menglu Cao, California State Polytechnic University, Pomona

The enormous advantages of active transportation lead the transportation research focus towards enhancing the walking and biking trips. The present study contributes to the current literature by determining the influential factors to the walking and biking travel frequency based on data obtained from the National Household Travel Survey (NHTS) California add-on survey. The study features some highlights. First, bivariate models were used to account for the common unobserved heterogeneity shared by the same persons and/or houses for the number of walking and biking trips. Second, endogeneity was explicitly considered due to the strong interdependency between walking and biking trips. Third, the bivariate normal distribution was applied to both household and person levels of random effects. Fourth, both variable importance ranking and correlation analyses were employed to determine the features to be fed into the models, which are different for each of the joint models. Fifth, to efficiently estimate the model parameters, a fast Bayesian inference approach, Integrated Nested Laplace Approximation (INLA) was used. Finally, distinct evaluation metrics were utilized for a comprehensive understanding of the model performance. The results illustrated that the models developed with endogeneity performed better than the those without endogeneity being included. Four influential variables, including mode to work by bicycle, public transit usage, count of household members, and multiple race responses, tend to have significantly significant impacts on walking and biking trips.

Xiaobing Li (xli158@ua.edu), University of Alabama
Praveena Penmetsa, University of Alabama
Jun Liu, University of Alabama
Alexander Hainen, University of Alabama
Shashi Nambisan, University of Nevada, Las Vegas

The transportation of natural gas often relies on pipelines which require constant monitoring and regular maintenance to prevent spills or leaks. Pipeline incidents could pose a huge adverse impact on people, the environment, and society. Numerous efforts have been invested to identify contributing factors to pipeline incidents so that countermeasures could be developed to proactively prevent some incidents and reduce incident severities or impacts. However, the countermeasures may need to vary for different incidents due to the potential heterogeneity between incidents, and such heterogeneity is likely related to the geology, weather, and built environment which vary across space and time domain. The objective of this study is to revisit the correlates of pipeline incidents, focusing on the spatial and temporal patterns of the correlations between natural gas pipeline incident severity and contributing factors. This study leveraged an integrated spatio-temporal modeling approach, namely the Geographically and Temporally Weighted Ordered Logistic Regression (GTWOLR) to model the natural gas pipeline incident report data (2010 - 2019) from the U.S. Pipeline and Hazardous Material Safety Administration. Text mining was performed to extract additional information from the narratives in reports. Results show several factors have significant spatiotemporally varying correlations with the pipeline incident severity, and these factors include excavation damage, gas explosion, iron pipes, longer incident response time, and longer pipe lifetime. Findings from this study are valuable for pipeline operators, end-users, responders to jointly develop localized strategies to maintain the natural gas distribution system. More implications are discussed in the paper.

Nazmul Arefin Khan, Dalhousie University
Muhammad Habib, Dalhousie University

This paper examines smartphone-based transport-support application usage for daily trip planning purposes, specifically, in the case of deciding departure time and trip destinations. Using a smartphone survey data, this study develops two latent segmentation-based random parameter logit (LSRPL) models that accommodate two layers of unobserved heterogeneity by introducing latent segments and random parameters within the modeling framework. It develops latent segment allocation models within LSRPL model formulations based on individuals’ socio-demographic characteristics, and probabilistically identifies two latent segments – tech savvy and non-tech savvy segments. The study exclusively estimates the influence of built environment and accessibility measures on transport-support application usage in trip planning purposes. For instance, people tend to use transport-support application more to decide their departure time while living in higher mixed land-use neighborhoods, whereas they have higher likelihood to use less transport-support applications to decide their trip destinations. Also, considerable heterogeneity is found during model analysis. For example, commuting by transit and a higher number of bikes in the household increases tech savvy individuals’ probability of higher TSA usage while deciding departure time. Non-tech savvy individuals on the other hand exhibit opposite relationships. Furthermore, heterogeneity is observed within both tech savvy and non-tech savvy segments that represent the preference variations among individuals with similar characteristics. Results of this study could be useful to evaluate ICT-based smart city policies that focus on improving quality and performance of overall transportation infrastructure.

Grigorios Fountas (G.Fountas@napier.ac.uk), Edinburgh Napier University
Achille Fonzone, Edinburgh Napier University
Adebola Olowosegun, Edinburgh Napier University
Clare McTigue, Edinburgh Napier University

This paper investigates the determinants of injury severities in single-bicycle and bicycle-motor vehicle accidents in Scotland using correlated random parameter ordered probit models with heterogeneity in the means. This modeling approach extends the frontier of the conventional random parameters by capturing the likely correlations among the random parameters and relaxing the fixed nature of the means for the mixing distributions of the random parameters. The empirical analysis was based on the publicly available database of police crash reports (STATS19) in the UK using data from accidents occurred on urban and rural carriageways between 2010 and 2018. The model estimation results show that various accident, road, location, weather, and driver or cyclist characteristics affect the injury severities for both categories of accidents. The heterogeneity-in-the-means structure allowed the incorporation of an additional layer of heterogeneity as the means of the random parameters are found to vary as a function of accident or driver/cyclist characteristics. The correlation of the random parameters enabled the identification of complex interactive effects of the unobserved characteristics captured by road, location and environmental factors. Overall, the determinants of injury severities are found to vary between single-bicycle and bicycle-motor vehicle accidents, whereas a number of common determinants are associated with different effects in terms of magnitude and sign. The comparison of the proposed methodological framework with less sophisticated ordered probit models demonstrated its relative benefits in terms of statistical fit and explanatory power as well as its potential to capture underlying heterogeneity to a greater extent.

Rohan Sood (soodrohan54@gmail.com), Imperial College London
Daniel Graham, Imperial College London
Ramandeep Singh, Imperial College London

Studies investigating the impact Daylight Savings Time (DST) has on road traffic casualty rates are inconclusive. This may be due to differences in geographical focus. Analysing road traffic accident data between 2005 and 2018, this paper employs a regression discontinuity design analysis to investigate the impact of DST on the total average road casualty and Fatal casualty rates across the UK. A novel approach is taken by dividing the UK longitudinally by Northing bands to investigate whether drivers in the north of Scotland are adversely impacted by DST. Overall, there is only a statistically significant change in average casualty rate when leaving DST in the autumn of -2.6%. There is no change in the average Fatal casualty rate at the UK level. There are observed differences in how DST impacts different parts of the UK at the longitudinal level. A clear band from the south of Scotland to the start of the Highland region has an estimated reduction of 27% in the average casualty rate when entering DST. However, other longitudinal results vary and in certain cases prompt further study. The methodology employed by this paper indicates that a longitudinal approach is recommended for future research into the effects of DST and on traffic casualties.

Sheikh Shahriar Ahmed (sheikhsh@buffalo.edu), University at Buffalo, SUNY
Jessica Cohen, Inferenx Labs, Inc.
Panagiotis Anastasopoulos, University at Buffalo, SUNY

Deer-vehicle collisions are a major type of crash in the United States. These collisions result in significant number of injuries and fatalities, and subsequent economic impacts, emphasizing the necessity to better understand factors contributing to such collisions. This study investigates deer-vehicle collisions in a bifurcated approach. Random parameters binary logit modeling framework with heterogeneity in means is employed to estimate two models: likelihood of witnessing a deer on the roadway, and likelihood of hitting a deer on the roadway. A database of crashes maintained by the Pennsylvania Department of Transportation from the year 2018 was used in the analysis. The analysis result from the first model revealed that it is more likely to witness deer in rural locations, in dark lighting conditions, and during deer breeding season when deer are more active in movement. Likewise, analysis results from the second model indicated that it is more likely for vehicles to hit deer in roads with a speed limit higher than 55 mph, and during deer breeding seasons. Female drivers were found to be more prone to hitting deer as compared to men. Additional insights were gained through the use of random parameters and heterogeneity in means approach, indicating the effect of unobserved heterogeneity in factors contributing to the likelihood of witnessing and hitting deer on the roadway.

Aupal Mondal, University of Texas, Austin
Chandra Bhat (bhat@mail.utexas.edu), University of Texas, Austin

Traditional multiple-discrete continuous choice models that have been formulated and applied in recent years consider a single linear resource constraint, which, when combined with consumer preferences, determines the optimal consumption point. However, in reality, consumers may face multiple resource constraints, such as those associated with time, money, and storage capacity. Ignoring such multiple constraints and instead using a single constraint can, and in general will, lead to poor data fit and inconsistent preference estimation, which can then have a serious negative downstream effect on forecasting and welfare/policy analysis. Unlike earlier attempts to address this multiple constraint situation, we formulate a new multiple-constraint (MC) multiple discrete continuous extreme value (MDCEV) model (or the MC-MDCEV model) that retains a closed-form probability structure and is as simple to estimate as the MDCEV model with one constraint. It is hoped that our proposed simple closed-form multi-constraint MDCEV model will contribute to a new direction of application possibilities and to new research into situations where consumers face multiple constraints within a multiple discrete-continuous choice context.