The Trip That Flips the Fear

Two denominators, two stories — both shown. The headline road rate (1.26 per 100M) is per vehicle-mile: every road death (drivers, passengers, motorcyclists, pedestrians, cyclists) over all miles driven. That gives ~79 miles per micromort. The drive-vs-fly bars instead use per passenger-mile occupant rates, the only fair way to compare a car seat with a plane seat — there car driving is ~189 miles per micromort. They are different questions, and the choice is itself the lesson.

The "240 miles per micromort" slogan is a kinder, older number. The figure you'll see quoted everywhere (~240–250 miles) implies a death only every ~240 million miles — a rate near 0.42 per 100M, about a third of the 2023 all-road rate. It reflects safer years and/or occupant-only counts. We don't repeat it as today's fact; we audit it against the raw number it disagrees with.

Per mile is one frame, not the only one. Per-mile, flying wins enormously. Per trip or per hour the gap narrows, because a flight covers vastly more miles per hour — and a fair door-to-door comparison should add the drive to and from the airport (which is, ironically, the riskiest leg of flying). The first calculator deliberately fixes distance and asks the per-mile question; it does not model the airport drive.

The per-hour numbers depend on an assumed speed — and it's shown. Per-hour risk is per-mile risk × a representative speed (mph). Savage reports per-mile and notes that "engineers might argue for … passenger-hours … to account for … the average speed of travel"; the per-hour reframing here is that transparent multiplication. The speeds (car ~35, bus ~30, train ~45, motorcycle ~35, air ~500 mph) are round representative values — the car figure is the softest (real trip speeds run ~25–70 depending on road), and the per-hour car number scales with it. But the conclusions don't: the air/ground speed gap is roughly 14×, so flying's per-mile lead over driving (~104×) must shrink to single digits per hour (~7×), and the slow, safe bus wins per hour for any speed between a city crawl and an intercity cruise.

Two eras, one ordering. The ruler table uses Savage's uniform 2000–2009 US rates because that is the one study measuring every mode the same way (onboard passengers only, one method). The car and air rates there (0.73 and 0.007 per 100M pax-mi) are a touch higher than the 2023 figures the first calculator uses (0.53 and 0.003) — both roads and skies got safer over two decades — but the ranking and the ruler-flip are unchanged. Modern transit totals quoted elsewhere (train ~0.19, bus ~0.09) look worse than Savage's onboard figures because they fold in trespassers, grade-crossing and platform deaths — people who were never riding. The honest denominator for "how risky is it to ride?" is the onboard-passenger rate, and that is what these bars use.

Why walking and cycling aren't on the bars. Savage leaves them out on purpose, and so do we: there is no reliable count of passenger-miles walked or cycled in the US — the denominator is a survey estimate, not a measurement, so any per-mile rate for them is built on sand. The best published US figures (Buehler & Pucher, Transport Reviews 2021: pedestrian ≈18, cyclist ≈10 deaths per 100M passenger-miles) are real but range-bearing, and — crucially — that acute crash rate ignores the large net health benefit of the exercise, which studies consistently find outweighs the crash risk in all-cause mortality. Put those two facts together and a bar labelled "walking" would mislead twice. The honest move is to name the gap rather than paper a false number over it.

The per-flight floor is a transparent free choice. Almost all aviation risk sits in take-off, climb, approach and landing, not cruise, so we add a representative 0.013 micromort fixed cost per flight. It barely moves a long trip, but it means that below roughly 2.5 miles the (absurd) flight would out-risk the drive — an honest boundary, not a hidden one.

The average hides enormous spread, and this page used to only say so. Until 2026-07-26 that sentence ended here, asserting an order of magnitude and checking none of it. It is now the section Whose average is it?, built from the raw FARS case files and FHWA's travel tables, where the order of magnitude turns out to be real (a motorcycle rider, more than twenty times the average per mile, both halves of the fraction published) and where the conditionings people actually invoke about themselves turn out to have no denominator at all. Commercial aviation's per-mile rate is separately dominated by rare catastrophic events, so single years are noisy. The ordering (drive >> fly per mile) survives every conditioning we can honestly apply, and the conditioning panel carries both complete FARS years so you can watch it survive the switch.

Three different ways of not knowing, and the page keeps them apart. A counted denominator is FHWA's traffic-counting programme, which is an estimate with real error but a measurement programme behind it; FHWA itself notes that travel on local roads and rural minor collectors "is estimated by the States based on a model or other means", so those cells are the softest of the counted ones. A surveyed denominator counts people rather than miles, under conditions chosen by the survey. None means nobody produces a figure. Separately, the alcohol numerator is itself modelled: NHTSA draws ten imputed blood-alcohol values for every driver never tested, and the published total is their mean. Our own count of deaths in crashes with a measured reading at 0.08 or above is far lower, because in 2024 roughly half of all road deaths were in crashes where no driver had a measured reading at all. Both figures are honest. They are not the same kind of thing, and the panel says which is which.