Category Archives: Recent research

Recent research

Transit Design in Response to a Global Pandemic

Leave a comment

Optimizing Operational Strategies for Mass Transit Systems in Response to a Global Pandemic

This is one of my COVID inspired research projects that was started in 2020.  The idea is that, in order to operate safety during a pandemic, transit agencies might have to adjust their operational strategies, in terms of service frequency and capacity.  The underlying tradeoff we are trying to explore here is that between the benefit of frequently testing drivers (as it reduces the transmission risk) and the cost of lowering the number of passengers allowed in buses, subject to the need to maintain certain safety standard, measured by infection risks. A novelty of the work is a physical model aiming to estimate infection risks based on vehicle size/type, service capacity and a few external risk factors.

The paper is currently under revision at Journal of Transportation Research Part A.   Please download a preprint here.

Abstract              This study analyzes the risk involved in riding various transit modes during and after a global pandemic. The goal is to identify which factors are related to this risk, how such a relationship can be represented in a manner amenable to analysis, and what a transit operator can do to mitigate the risk while running its service as efficiently as possible. The resulted infection risk model is sensitive to such factors as prevalence of infection, baseline transmission probability, social distance, and expected number of human contacts. Built on this model, we formulate, analyze and test three versions of a transit operator’s design problem. In the first, the operator seeks to jointly optimize vehicle capacity and staff testing frequency while keeping the original service schedule and satisfying the infection risk requirement. The second model assumes the operator is obligated to meet the returning demand after the peak of the pandemic. The third allows the operator to run more than one transit line and to allocate limited resources between the lines, subject to the penalty of unserved passengers. We find: (i) The optimal profit, as well as the testing frequency and the vehicle capacity, decreases when passengers expect to come in close contact with more fellow riders in a trip; (ii) Using a larger bus and/or reducing the testing cost enables the operator to both test drivers more frequently and allow more passengers in each bus; (iii) If passengers weigh the risk of riding bus relative to taxi, a higher prevalence of infection has a negative effect on transit operation, whereas a higher basic transmission probability has a positive effect; (iv) The benefit of improving service capacity and/or testing more frequently is limited given the safety requirement imposed. When the demand rises beyond the range of the capacity needed to maintain sufficient social distancing, the operator has no choice but to increase the service frequency; and (v) In the multi-line case, the lines that have a larger pre-pandemic demand, a higher penalty for each unserved passenger, or a greater exposure risk should be prioritized.

Recent research

An Efficiency Paradox of Uberization

Leave a comment

The main finding is that e-hail (e.g., Uber and Lyft) may not scale as well as street-hailing taxi.  In other words, a thicker market may help improve the performance of taxi more than that of e-hail. We develop a physical model to describe the matching process of both modes. Using the model, we then derive the production function and measure the returns to scale in the matching process.  It indicates that taxi has returns to scale of  2 (implying increasing returns to scale) whereas e-hail has returns to scale of 1 (constant returns to scale).   Empirical data collected in Shenzhen, China largely confirm the theory.

While this paper has yet to be published, I must have spent more time writing and revising it than any of my other papers.  Hongyu convinced me to write it  for a general journal, which, with the benefit of hindsight, might have caused us to oversell an otherwise fine idea.   That being said, I did learn a lot from the process, and I still think it is one of my better works.   The preprint is here, and the abstract follows.

Uberization promises to transform society based on an intuitive proposition: Advanced peer-to-peer matching guarantees greater overall efficiency. Here we show a paradox challenging this proposition in uberized ride-hail service, known as e-hail. By analyzing hundreds of local markets in Shenzhen, China, we discover e-hail is outperformed—in terms of wait time and trip production—by taxis hailed off street in areas with high densities of passengers and drivers. This paradox arises because a quicker match does not always expedite and enhance a service. On the contrary, it can induce competition that undermines the network effect, making a passenger less likely to benefit from more drivers, and vice versa, in e-hail than in taxi service. Consequently, simply attracting more users may not improve e-hail’s efficiency (defined as trip production at a given density of passengers and drivers), because its competitive edge diminishes with scale. The finding implies uberization has a limited impact on efficiency and is unlikely to create a “winner-take-all” in transportation

Recent research

A-PASS for Travel Demand Management

Leave a comment
Auction-Based Permit Allocation and Sharing System (A-Pass) for Travel Demand Management, co-authored by Ruijie Li and Xiaobo Liu (both at Southwest Jiaotong University, China).

This is a follow up to another paper related to the mechanism design problems arising from ridesharing.  In this paper, we try to show the promise of integrating ridesharing with quantity-based travel demand management.  One of the main insights is that by auctioning out permits (e.g., to use a road facility), we can eliminate the deficits that are otherwise unavoidable in classical  Vickrey-Clark-Gloves mechanism.  The paper just came out in Transportation Science. You may read the abstract below and download a preprint here.

We propose a novel quantity-based demand management system aiming to promote ride-sharing. The system sells the permit to access a facility (conceptualized as a bottleneck) by auction but encourages commuters to share the permits with each other. The permit is classified according to access time and the commuters may be assigned one of the three roles: solo driver, ride-sharing driver, or rider. At the core of this auction-based permit allocation and sharing system (A-PASS) is a trilateral matching problem (TMP) that matches permits, drivers and riders. We formulate TMP as an integer program, and prove it can be reduced to an equivalent linear program. A pricing policy based on the classical Vickrey-Clark-Gloves (VCG) mechanism is proposed to determine the payment for each commuter. We prove, under the VCG policy, different commuters will pay exactly the same price as long as their role and access time are the same. We also show A-PASS can eliminate any deficit that may arise from the VCG policy by controlling the number of shared rides. Results of numerical experiment suggest A-PASS strongly promote rider-sharing. As ride-sharing increases, all stake holders are better off: the ride-sharing platform receives greater profits, the commuters enjoy higher utility, and the society benefits from more efficient utilization of infrastructure.

Recent research

A Physical Model of Street Ride-Hail

Leave a comment

The work was initiated by my former student, Hongyu Chen, as part of his PhD research.  He wanted to build a “physical model” of street-hail taxi matching that can be calibrated and validated with real data (taxi trajectory).  This is very difficult because passenger wait time cannot be directly observed in taxi trajectory data. Hongyu came up with a rather clever and elaborated method to accomplish just that. Read the abstract below and download the paper here.

In this study, we show that the passenger-driver matching process in street ride-hail is dictated by the physical limitation of a passenger’s average eyesight and the preference of cruising taxi drivers for certain locations. Together, these two spatiotemporal features, called effective hail distance (EHD) and local area attractiveness (LAA) respectively, define the number of vacant taxis that a passenger can reach, and accordingly the distribution of her waiting time. To calibrate the waiting time distribution, we extract maximum possible waiting times from taxis GPS trajectory data, by tracking the movements of vacant taxis cruising around a pickup location. Then we prove that, for a given EHD, the extracted maximum possible waiting time follows the same distribution as passenger waiting time. The proposed matching mechanism, along with the novel calibration method, leads to a general model of street ride-hail that can produce reliable estimates of passenger waiting time under a wide variety of market conditions. Moreover, the matching process in the phone-based ride-hail is shown to be a special case of the proposed model, when EHD approaches infinity. This result lays the foundation for understanding and comparing the performance of ride-hail services. It can also help address regulatory and operational questions facing key stake holders in this industry.