The Hidden Gridlock: Why Traffic Models Keep Failing Modern Cities

[SECTION: The Prediction Problem] Every major city on earth has a traffic model. Some of them cost tens of millions of dollars to build, require supercomputers to run, and employ teams of engineers who spend careers refining their assumptions. And nearly all of them, with remarkable consistency, get the future wrong. The failure is not a matter of bad data or insufficient computing power. It is something more fundamental — a mismatch between the way transportation planners think about movement and the way movement actually works in complex, adaptive human systems. Cities are not machines. They are ecosystems. And ecosystems have a habit of responding to interventions in ways that confound the people who designed the intervention in the first place. [SECTION: Induced Demand and the Expanding Highway] The most well-documented example of this mismatch is induced demand — the phenomenon whereby adding road capacity generates new traffic rather than relieving existing congestion. The effect has been studied extensively since the 1990s, and the evidence is robust enough that most transportation economists now treat it as settled. University of Toronto research on induced demand Yet highway expansion projects continue to be justified with traffic models that systematically undercount it. [STAT: 100% | approximate increase in vehicle miles traveled for every 10% increase in highway lane miles, per multiple peer-reviewed studies] This is not a rounding error. It is a structural blind spot baked into the modeling process itself, partly because the political economy of transportation funding rewards projections that justify construction, and partly because the feedback loops between supply, behavior, and land use operate across timescales that standard models compress or ignore entirely. When a new interchange opens, it does not simply redistribute existing trips more efficiently. It changes where people choose to live, where employers choose to locate, which businesses become viable, and what kinds of trips feel worth taking. The road reshapes the city, and the reshaped city generates the very congestion the road was meant to prevent. This cascade typically unfolds over five to fifteen years — long enough that the original planners have moved on, and long enough that the connection between cause and effect becomes politically invisible. [SECTION: The Transit Paradox] Public transit systems face a mirror-image version of the same problem. Ridership models for new rail lines routinely overestimate usage in the early years and underestimate the longer-term network effects that emerge once a line becomes embedded in the city's spatial logic. The opening of a single subway extension can shift real estate values across an entire corridor, attract density that was not anticipated in the original environmental review, and eventually generate ridership that the model would have called implausible. [STAT: 20-30% | typical range by which major transit ridership forecasts have historically overestimated first-year usage, according to analyses by transport researcher Bent Flyvbjerg] The optimism bias runs in the opposite direction from highway modeling, but the underlying error is the same: treating a transportation intervention as a static addition to a fixed system rather than a perturbation in a dynamic one. What makes this particularly consequential is the funding structure. Transit projects are approved based on projected ridership, which means that early underperformance creates political vulnerability precisely during the period when the network effects have not yet had time to materialize. Lines get underfunded, service gets cut, ridership falls further, and a self-reinforcing decline begins — a cascade that the original model could not have predicted because it was never designed to model feedback. [SECTION: Micromobility and the New Variables] The arrival of e-bikes, scooters, and ride-hailing platforms has added new layers of complexity that existing models are almost entirely unprepared to handle. These modes do not simply fill gaps in the transportation network. They interact with transit, reshape short-trip behavior, affect parking demand, and alter the economics of car ownership in ways that vary dramatically by neighborhood density, demographics, and climate. Some cities have found that e-bike adoption reduces car trips more effectively than any infrastructure investment they have made in decades. Others have found that ride-hailing increases vehicle miles traveled by adding deadhead miles — drivers circling without passengers — that no one had accounted for. Schaller Consulting report on ride-hailing and congestion The difference between these outcomes depends on local conditions that are extraordinarily difficult to parameterize in advance. [SECTION: Toward Adaptive Planning] The answer is not better models, exactly — or not only that. It is a different relationship between models and decisions. Transportation planning has historically treated the model as an oracle: you run it, you get a number, you build the thing. A systems-science approach treats the model as a hypothesis generator, something to be tested against real-world feedback and revised continuously as the city responds.

A handful of cities are beginning to move in this direction, building monitoring infrastructure that allows them to track behavioral responses to interventions in near real time and adjust service, pricing, and capacity accordingly. It is slower and less politically satisfying than a ribbon-cutting. But it is considerably more honest about what cities actually are — living systems that talk back.