Everyone picks the fastest route — and it stops being fastest
Traffic is a textbook example of a system where individually rational choices do not add up to a collectively rational outcome. Every driver simply wants the fastest route available to them right now, and every route gets slower the more cars use it. When enough drivers pile onto the route that currently looks fastest, it becomes just as slow as the alternatives, and drivers redistribute until no one can do better by switching — a self-organising, decentralised equilibrium reached with no central planner and no communication between drivers.
Wardrop's equilibrium, and why it is not optimal
John Wardrop formalised this condition in 1952: in equilibrium, every route actually being used between an origin and a destination has equal travel time, and that time is no greater than any unused alternative would provide. This is the traffic-network analogue of a Nash equilibrium — no single driver benefits from unilaterally switching routes — and it is what a road network naturally converges to when drivers are left to choose freely.
The catch is that a Wardrop equilibrium is generally not the system optimum — the routing pattern that minimises the sum of everyone's travel time. Every driver who joins a congested route imposes a small additional delay on every other driver already on it, a cost the joining driver does not pay and typically does not even see. Tim Roughgarden and Éva Tardos formalised how large this gap can get with the concept of the price of anarchy — the ratio between the total travel time under selfish routing and under the true optimum — showing it can be substantial even in simple networks with standard congestion functions.
Braess's paradox: more road, more delay
Dietrich Braess demonstrated in 1968 one of the most counterintuitive results in network theory: adding a new link to a road network can increase everyone's travel time at the new selfish equilibrium, even though the new link is, in isolation, strictly faster than any existing route it replaces. The mechanism is that the new link is so individually attractive that every driver rationally chooses to use it, funnelling traffic onto a shared bottleneck downstream that did not exist as a shared constraint before — the network's Wardrop equilibrium shifts to a worse configuration overall, even though no individual driver made an irrational choice. The effect has been documented in real cities, including cases where closing a congested road unexpectedly improved traffic flow.
Wardrop equilibrium: every USED route between O and D has equal travel time,
no worse than any unused alternative
system optimum: the routing pattern minimising TOTAL travel time
(generally requires some drivers to accept a
personally slower route for the group's benefit)
price of anarchy = total time at Wardrop equilibrium
/ total time at system optimum (always ≥ 1)
Tolling as a Pigouvian correction
Congestion is a textbook negative externality — each driver's marginal trip imposes a real cost on everyone else that the driver does not pay. A Pigouvian toll, priced to match the external delay a marginal driver imposes on the route, internalises that externality: once the toll is added to the private cost of a congested route, the route no longer looks artificially attractive relative to its true social cost, and the driver's self-interested choice starts to align with the system-optimal routing pattern rather than diverging from it. This is the theoretical basis behind real congestion-pricing programmes — London's congestion charge, Singapore's Electronic Road Pricing, and New York's 2025 congestion pricing scheme — and it is also the mechanism that can be used to defuse a Braess's-paradox situation by discouraging exactly the traffic pattern that made the new link counterproductive.
Frequently asked questions
What is a Wardrop equilibrium in traffic?
It is the state where every route actually being used between an origin and destination has the same travel time, and that time is no worse than any unused alternative route would offer. No individual driver can improve their own trip by unilaterally switching routes, which makes it the traffic-network analogue of a Nash equilibrium — but it is generally not the routing pattern that minimises total travel time across everyone.
How can adding a new road make traffic worse for everyone?
This is Braess's paradox: a new link can be so individually attractive that every driver rationally chooses to use it, but doing so shifts the whole network into a new selfish equilibrium with a higher total travel time than before the road existed — even though the new route was, in isolation, an improvement. The effect has been observed and reversed in real cities, including cases where closing a road reduced congestion.
How does a congestion charge actually reduce travel time?
A congestion toll is a Pigouvian tax: it makes each driver pay a price roughly equal to the extra delay their trip imposes on everyone else, a cost that is otherwise invisible to that driver. Once that externality is priced in, the driver's private incentive lines up with the system-wide optimum, which is exactly what marginal-cost tolling is designed to achieve — and it is the mechanism behind real programmes like London's congestion charge and Singapore's Electronic Road Pricing.
Try it live
Everything above runs in your browser — open Congestion Pricing and change the parameters while it is running. Nothing is installed, nothing is uploaded, the whole model lives in one tab.
▶ Open Congestion Pricing simulation