The AI Journal

Central London has charged drivers to enter its centre every weekday since February 2003. Stockholm has done the same since 2007. Singapore has priced its roads in some form since 1975, moving from a manned Area Licensing Scheme to fully electronic tolling in 1998 and adjusting rates ever since without provoking the crisis that ends most such schemes elsewhere. Measured against the size of the problem congestion pricing is meant to solve, this is a strikingly short list of durable examples. William Vickrey set out the efficiency case in 1963: charge drivers something close to the cost their trip imposes on everyone else on the road, and congestion falls, travel times become predictable, and the money raised can fund the alternatives that make the charge bearable. Economists have not seriously disputed this since. What they have had to explain instead is why, sixty years on, almost no city has done it. Edinburgh put a scheme to a public vote in 2005 and 74 per cent voted no. Greater Manchester put a larger one to a vote in 2008, tied to a package of transit investment partly funded by the charge, and 79 per cent voted no in every one of its ten boroughs. New York City abandoned a 2007 proposal before it reached a vote at all, spent most of the following two decades revisiting the idea, and finally began tolling a much-reduced version in January 2025 — after which a federal transportation secretary tried to terminate it by letter, and a court had to rule the attempt unlawful before the programme could continue.

The standard account of these defeats treats them as price resistance: drivers dislike paying for what was free, and no design fixes that. A second account treats them as regressivity: a flat daily charge falls harder on lower earners, so opposition tracks income. Both have some truth in them, and Kenneth Small’s 1992 proposal for recycling revenue through direct rebates and reduced payroll or sales taxes was built explicitly to answer the regressivity objection. But neither account explains the pattern in the results that did happen. London’s charge was approved by no popular vote at all; the mayor who introduced it had campaigned for it and was re-elected afterwards. Stockholm’s did go to a referendum, in September 2006, after a seven-month trial had already been running since January that year — and the charge, having been felt rather than merely promised, passed. Manchester and Edinburgh asked voters to approve a charge before any of it existed and before any of the promised transit money had been spent.

Michael Manville and David King’s 2013 analysis names the mechanism directly: congestion pricing is a credible-commitment problem before it is a pricing problem. A government proposing a toll is asking residents to bear a certain, immediate cost — money out of pocket every weekday — in exchange for an uncertain, future benefit: less traffic, better buses, lower charges over time as targets are met. Nothing binds the government to deliver the second half of that bargain once the first half has already been collected. Voters know this, and rationally discount promises about revenue use to something close to zero. A scheme that instead demonstrates its own case — a trial period long enough for commuters to experience shorter journeys, buses that arrive on their published schedule, revenue that is visibly spent rather than merely allocated on paper — converts an unverifiable promise into an observed fact before it asks for a permanent yes. Jonas Eliasson’s account of the Stockholm trial and the follow-up study by Geertje Schuitema, Linda Steg and Sonja Forward both find the same shift: public support for the charge was measurably higher after residents had lived with it than it had been in any survey taken beforehand, and the improvement tracked people’s own reports of shorter, more predictable trips rather than any change in their stated views about fairness or regressivity. The dislike of paying did not go away. It stopped being decisive once it was weighed against a benefit people had actually banked rather than merely been promised.

David King, Manville and Donald Shoup’s earlier study of the same problem adds a second, related finding: support for congestion pricing tracks not just whether revenue returns, but whether the return is visible to the people paying, delivered close to where they live, and hard for a future administration to redirect. A charge whose revenue disappears into a general transport fund invites exactly the suspicion a demonstration period is built to dispel. London’s revenue is earmarked by statute for the city’s own transport authority; the charge has been adjusted and extended several times since 2003 without triggering the kind of referendum defeat that ended the Manchester and Edinburgh proposals, because there was never a moment when voters were asked to trust an unverified promise about where the money would go. New York’s programme, though never put to a popular vote, followed the same instinct: state legislation routed its toll revenue into a dedicated capital account for the transit authority’s subway and rail improvements rather than the general budget, so that money collected from drivers converts directly into a published, trackable list of projects rather than an unaccounted-for pool a future administration could quietly redirect.

The strongest objection to this account is that trials and earmarking are themselves cheap, and that a government determined to avoid a charge altogether can simply refuse to run one long enough to convince anyone. A short, favourably timed pilot could manufacture the appearance of a credible commitment without the substance of one — running it during a period of unusually light traffic, for instance, or before an unrelated transit improvement that was already scheduled independently. This is a real risk, and it means a trial’s evidentiary value depends on details a hostile government has every incentive to distort: its length, its timing relative to other changes, and whether the data collected during it are published in a form outsiders can check. But this narrows the design problem rather than eliminating it. It argues for trials specified and audited by a body other than the one proposing the charge, not for abandoning the trial-before-vote sequence in favour of the up-front referendum that has failed almost everywhere it has been tried. The alternative to an imperfect demonstration is not a perfect one; it is asking voters to take the entire bargain on faith, which is precisely the request the record shows they refuse.

Congestion pricing does not fail because drivers will not pay for roads. Singapore’s drivers have paid for forty years; London’s have paid for over twenty. It fails when a government asks to be trusted before it has shown anything, and succeeds, however narrowly, once it has let the toll make its own case first.