The Strategy of Waiting
Whether it is a ridesharing driver seeking a fare at a crowded airport or a patient hoping for a life-saving organ, modern life depends on queues and mechanisms that match people with limited, time-sensitive resources. Queueing theory, one of the earliest pillars of industrial engineering and operations research, began in the early 20th century with studies of telephone congestion and has since grown into a foundational framework for analyzing complex service operations. Yet queues often fail to operate as intended. When individuals decline offers, waiting for a better ride or hoping for a more suitable organ, the effects ripple outward, slowing matches, extending wait times, and leaving valuable resources unused.
A new study by researchers at UC Berkeley IEOR and Columbia University reveals why strategic behavior can be so costly and introduces a mathematical framework for evaluating its effects. The paper, “An Equilibrium Solver for a Dynamic Queueing Game,” presented at the 2025 ACM Conference on Economics and Computation (EC’25), offers one of the first tools capable of capturing complex strategic behaviors in real-world queueing environments and quantifying how those decisions influence overall system performance.
When waiting becomes strategic
A queue may look straightforward, such as the line of ridesharing drivers waiting outside an airport. At first glance, it seems like a simple first-come, first-served process. But once a driver receives a trip, the decision becomes strategic. They weigh the destination, earnings, time of day, their queue positions and even the behavior of other drivers. One person’s decision to accept or decline a trip reshapes the opportunities available to everyone else, transforming what appears to be a simple line into an interconnected system of strategic choices.
“What makes these systems challenging to analyze is that one person’s decision changes the opportunities available to everyone else,” said Denise Cerna, a doctoral student in IEOR and lead author of the study. “Those interdependencies shape the entire stochastic dynamics of the queue, and understanding them is essential for designing policies that don’t inadvertently create delays or waste.”
These decisions carry real consequences. In the U.S. organ transplantation network, roughly one in four deceased-donor kidneys goes unused.1 Even after years on a waitlist, patients and clinicians sometimes reject kidneys because organ quality can vary significantly, and accepting a poorer-quality organ may reduce long-term survival. Strategic declines—rational from an individual’s perspective—can delay matches and, in some cases, lead to the discard of organs that could have benefited someone else.
Similar behavior appears in ridesharing: at airports like San Francisco International, trip completion rates have fallen to around 60% since platforms began displaying destinations and earnings in advance, unintentionally encouraging drivers to wait for higher-value trips. Every declined offer forces the system to search again, reassign the request—slowing both drivers and passengers caught in the queue.
Decades of queueing research have produced powerful methods for analyzing system performance under structured assumptions. Yet capturing strategic behavior in dynamic queues—where agents constantly arrive, depart, and choose whether to accept or decline offers—remains analytically difficult in general. To address this challenge, Cerna and her advisor, Assistant Professor Chiwei Yan, together with Columbia University Assistant Professor Hongyao Ma, developed a solver that models queues as they operate in practice: dynamically, with heterogeneous items and forward-looking strategic agents whose decisions influence every subsequent step.
Their method computes equilibrium strategies—capturing agents’ decisions to accept, reject, join, or leave queues as functions of both queue length and queue position—across a wide range of dispatch rules, providing researchers with a general-purpose tool for studying strategic behavior in dynamic, real-time environments.
Testing in real-world systems
When tested on real ridesharing data from Chicago O’Hare International Airport, the solver replicates equilibrium outcomes predicted by simpler dispatch policies, validating its accuracy in settings that earlier
theoretical work has successfully characterized. Its most significant contributions, however, arise in environments that prior analyses could not address—particularly richer policies that adapt to stochastically evolving queue lengths, which are highly relevant for real-world implementation on ridesharing platforms. The framework can also evaluate policies that broadcast requests to multiple drivers simultaneously, a practice increasingly adopted by modern platforms. Notably, this capability allows the solver to elucidate competitive dynamics that emerge when drivers operate across multiple platforms or when riders submit requests on more than one app.
While ridesharing provides a vivid illustration, the solver’s applicability extends to a wide range of domains, including the allocation of deceased-donor organs to patients on transplant waitlists, as well as any socio-technical system that operates as a dynamic queue in which individuals or institutions strategically accept or decline offers that arrive online. In each case, delays and inefficiencies arise not merely from resource scarcity, but from the layered and interdependent decisions of participants embedded in these processes. By offering a method that captures these interactions with far greater fidelity, the UC Berkeley–Columbia research team establishes a new foundation for evaluating and improving policies in sectors where misaligned incentives can carry substantial social or economic costs. Policymakers can use the solver to test how proposed reforms may influence participation, waiting times, and resource utilization before those changes take effect.
“Ultimately, our work charts a path toward service systems that allocate resources more efficiently, adapt to strategic behavior rather than ignore it, and better serve the people who rely on them,” said Professor Chiwei Yan.