Is Zero Reasonable? A Multidisciplinary Framing of Traffic Deaths

The whole debate snaps into focus the moment you pull apart the readings hiding inside "reasonable to expect."

• Prediction. Will the annual toll ever print as 0? A claim about the future behavior of a chaotic, 230-million-driver system.
• Feasibility. Could it, in principle — given the right roads, vehicles, speeds, and rules? A claim about what is physically and institutionally achievable.
• Commitment. Should we treat zero as the organizing goal even if we never literally touch it? A claim about how to set targets and assign blame.

Most "zero" arguments are people answering different questions and mistaking it for disagreement. The realist says we'll never predict zero; the advocate says we must commit to zero. Both can be right at once, because they aren't making the same kind of statement. Hold that distinction and the rest of this essay is just filling in what each discipline contributes to which question.

Injury epidemiology reframed road death decades ago, and that reframing is the foundation everything else rests on. To a public-health researcher a crash isn't bad luck — it's a preventable injury event with identifiable causes and dose-response curves, no different in kind from cholera or lung cancer. William Haddon Jr., the first head of what became NHTSA, built the Haddon Matrix: a grid crossing three phases (pre-crash, crash, post-crash) against three factors (human, vehicle, environment). Overnight a single death had nine cells of possible intervention instead of one — and only one of them was "the driver screwed up."

That quietly dismantled the dominant theory of the 20th century — that crashes come from "the nut behind the wheel," so the fix is better nuts. The epidemiological frame says you cut the disease burden by changing the agent (crash energy) and the environment, because you will never finish reforming 230 million humans. Seatbelts, airbags, guardrails, breakaway sign posts: all of them came from people who stopped asking whom to blame and started asking how to keep kinetic energy from reaching a body.

Sweden pushed the public-health frame to its logical end in 1997 and called it Vision Zero. Its engineering premise has two halves: humans will make mistakes, and human bodies tolerate only so much force — an unprotected person struck at 30 km/h usually lives, at 50 km/h usually dies. So the design goal isn't error-free drivers; it's a system where an ordinary mistake doesn't convert into a corpse. Speeds get matched to the crashes a road actually permits; conflict points are engineered out; kinetic energy is held below what a body can survive.

This is defense in depth — James Reason's "Swiss cheese" model, borrowed from aviation and nuclear safety. No single layer (driver, vehicle, road, speed limit, trauma response) has to be perfect; the holes just can't line up. The genuinely radical move is the reassignment of responsibility: when someone dies, the Safe System asks what the designers could have done, not only what the victim did wrong. Through this lens, near-zero stops looking utopian and starts looking like an engineering target several countries are visibly closing in on.

The numbers complicate the optimism in two directions at once. First, the denominator decides the story: the fatality rate per 100 million vehicle-miles has collapsed over a century even as the raw count plateaued, because Americans drive vastly more. "Huge progress" and "stuck for a decade" are both true — they're just different fractions. Second, death is a rare event smeared across hundreds of billions of trips a year. You can drive the per-trip probability arbitrarily low and still expect a non-trivial count, the way a tiny per-flight risk still yields crashes at scale. Pushing a Poisson process to a hard, permanent zero is a categorically stronger claim than pushing it very low.

Which is where the favorite analogy — "commercial aviation is basically at zero, so why not roads?" — both inspires and misleads. US airline travel really has gone years with near-zero passenger deaths per billion miles. But aviation is a closed system: a few tens of thousands of certified professional crews, redundant aircraft, rigid procedure, total traffic control, and a no-fault investigation culture that feeds every incident back into design. The road network is an open system — ~230 million amateur operators, no recurrent certification, almost no redundancy, machines sharing space with toddlers and deer. Aviation proves near-zero is possible; it also prices the ticket, and the price is a degree of control roads will never have.

Welfare economics asks the uncomfortable question out loud: is zero the efficient target at all? In the textbook model, the marginal cost of preventing one more death eventually exceeds the marginal benefit, so the "optimal" number of crashes is greater than zero. Economists make this tractable with the value of a statistical life — the US Department of Transportation uses roughly $12–13 million — which lets a guardrail's cost be weighed against the deaths it prevents. By that arithmetic, spending without limit to save the last life is irrational.

Vision Zero's deepest move is to reject the premise of that whole calculation. It refuses to treat a human life as one priced input trading against travel time and construction budgets. Economics can still do enormous work inside the commitment — sequencing which countermeasures buy the most safety per dollar — but it cannot, by itself, settle whether a fatality is an acceptable line item. That's a values question wearing a spreadsheet. (Economics also flags a feedback trap: safety gains can be partly spent as extra speed or risk — the Peltzman effect, or risk compensation — which is the handoff to psychology.)

Strip away the engineering and a moral question is left standing: who decides what risk is tolerable, and for whom? A utilitarian can live with a non-zero optimum if total welfare is maximized. A deontologist — and Vision Zero is essentially deontological — answers that using some people's deaths as the cost of others' convenience treats them as means, not ends. Hence Tingvall and Haworth's founding axiom: it can never be ethically acceptable that people are killed or seriously injured while moving through the road system.

Distributive justice sharpens the knife. Road deaths aren't shared evenly: pedestrians and cyclists, rural drivers, low-income communities, and Black and Native American Americans are over-represented in the toll. "Acceptable risk" is always acceptable to someone — usually not the person bearing it. It's also why the popular trolley-problem framing of self-driving cars is mostly a distraction: the real ethics of automation lives in mundane, structural choices — how fast to deploy, who's exposed to immature systems, how liability shifts — not in lifeboat dilemmas a car meets once a decade.

NHTSA's much-quoted line that ~94% of crashes involve human error is true and misleading in the same breath: an accurate tally of the last link, and a quiet alibi for everything upstream that made the error lethal. Behavioral science explains the proximate failures — speed, impairment, distraction, fatigue, and the way people partly spend safety gains on faster, looser driving. But the structural story is louder. The United States kills its people on the road at roughly two to four times the rate of peers like Sweden, the UK, or Japan — not because Americans are uniquely reckless, but because of exposure and design: deep car dependency, more miles driven, fast multi-lane "stroads" knifing through places people walk, heavier and taller vehicles, induced demand, and land-use patterns that turn every errand into a high-speed trip.

That's the hope hiding inside the bad news. If the gap with peer nations were a matter of national character, it would be hopeless. Because it's a matter of roads, speeds, vehicles, and rules, it is — in principle — engineerable down.

If human error is the dominant last link, automation is the obvious lever. Today's ADAS features — automatic emergency braking, lane-keeping — already measurably cut specific crash types, and mandating them is among the highest-yield moves on the table. Fully self-driving vehicles promise to remove most of the human-error contribution outright.

But technology reframes "zero" more than it delivers it on any near horizon. Autonomous systems trade familiar failures for new ones: the long tail of bizarre edge cases, brittle behavior in mixed human-and-robot traffic, cybersecurity, and accountability that migrates from drivers to manufacturers and code. Fleet turnover takes decades — a car bought today may still be on the road in 2045. The honest projection is that automation can plausibly push deaths toward zero over a multi-decade horizon while changing what "expect" even means, because the question becomes how safe we require the machines to be before we trust them — which loops straight back to ethics and economics.

Lay the three readings side by side and the question mostly answers itself.

• As a forecast: no. A literal, permanent count-zero isn't a reasonable prediction on any near horizon. Rare events across hundreds of billions of trips, an open system of amateur operators, and decades of fleet turnover make a hard zero the wrong kind of claim to bet on.
• As an engineering target: largely yes — near-zero is reachable. Peer countries and specific Safe-System cities have driven deaths down sharply with tools we already own: speed management, separation of users, vehicle standards, automatic braking. The remaining gap is political and infrastructural, not a law of physics.
• As a moral commitment: yes — and arguably the only defensible stance. The alternative is to write down the number of preventable deaths you'd accept, and no official will say it aloud, because it can't be said.

The cleanest reframing is to stop asking "can we hit zero?" and start asking "is any death we could have designed out acceptable?" Those are different questions, and the second is the one Vision Zero actually poses. "Reasonable to expect" then resolves three ways at once: reasonable to predict — no; reasonable to pursue — yes; reasonable to organize around — the most useful reading of all. The disciplines don't really disagree. They're answering different questions, and a serious conversation starts by saying which one you mean.