Wheels Down, Still Late

Summary

A US flight is recorded as on time if it reaches its gate within fifteen minutes of schedule. Not if it lands on time — if it parks on time. The interval between the two, taxi-in, is on every flight record the DOT publishes and appears in no consumer product. Across 17,070,588 completed arrivals into the 64 airports FlyerIntel covers, taxi-in averaged 9.02 min, arrivals-weighted. The average understates how it is distributed: the median arrival taxied 7 min, one in ten taxied 16 min or more, and one in a hundred taxied 38 min or more. And it is doing measurable work in the on-time figures. 196,683 arrivals — 5.4% of every delayed arrival, roughly one in 18 — had their wheels on the ground at or before the arrival time printed on the ticket and were still recorded as delayed, because they were still taxiing when the clock ran out.

The US Department of Transportation measures arrival at the moment the aircraft parks and the parking brake is set, not at the moment the wheels touch the runway. Every on-time percentage published in America — by the DOT, by airlines, by every booking site that quotes one, and by FlyerIntel — is a gate measurement. That is the right definition, because a passenger cannot leave an aircraft that is still moving. But it means a quantity almost nobody has heard of sits inside every one of those numbers: taxi-in, the interval between wheels on and gate on, reported by the operating carrier on every single flight. It is in the public record for every flight behind this report. It is on no ticket, in no schedule, and in no airline's marketing. It is also one of the few parts of a journey a passenger watches happen through a window while being told nothing about it.

Because arrival delay is measured at the gate and taxi-in is measured separately, the two can be subtracted, and the moment of touchdown can be placed against the arrival time on the ticket. Doing that for every completed arrival into a covered airport produces a set of flights that should not exist and does: 196,683 of them, 5.4% of every arrival recorded fifteen or more minutes late, whose wheels were already on the ground at or before the scheduled arrival time. They were not late in the air. They were late on the ground, at the destination, after landing. Those flights taxied an average of 38.1 min and were recorded a median of 21 min late, so this is not a rounding artefact around the fifteen-minute line — it is more than half an hour of taxiing, charged to a flight that had arrived. Loosen the test from the scheduled gate time to the DOT's own fifteen-minute tolerance and the count rises to 874,075, 24.1% of all delayed arrivals: nearly a quarter of America's arrival delay is measured after the aeroplane has finished flying. The strict figure is the safe one and it is deliberately conservative, because the scheduled gate time already contains an allowance for taxiing that this comparison hands back to the flight.

Every column is an arrivals measure: completed flights that landed at this airport. The last column is the share of the airport's arrivals recorded fifteen or more minutes late whose wheels were down at or before their scheduled arrival time.

ORD (Chicago) is the slowest of the 64 airports measured at getting an arriving passenger to a gate: 17.68 min on average across 899,457 arrivals, with 29.2% of them taxiing twenty minutes or more. SJC (San Jose) manages 4.14 min. The gap between them, 13.5 min, is longer than the entire taxi-in at 62 of the 64 airports in the table. How that spread is summarised changes the answer, so it is worth stating plainly which one is used here. Averaging the 64 airport averages together, one airport one vote, gives 7.37 min. Weighting each airport by the arrivals it actually handled gives 9.02 min — 1.65 min more, because the slow airports are disproportionately the big ones and a passenger is far more likely to be landing at one of them. Every network figure in this report is arrivals-weighted.

Arrivals-weighted across every covered airport, taxi-in is lightest at 23:00 at 8.12 min and heaviest at 18:00 at 10.16 min. A 2.0 min swing across the day is real but modest, and on its own it would be a dull finding. The interesting part is what is moving. The median arrival taxies between 6 min and 8 min whatever the hour — the typical experience is close to constant from breakfast to midnight. What changes is how often the taxi goes badly wrong: the share of arrivals taxiing twenty minutes or more runs from 3.8% at 12:00 to 9.6% at 18:00, 2.5 times as often. Arriving in the evening bank does not make your taxi a little longer. It multiplies your chance of the taxi that costs you a connection by about 2.5.

Scheduled arrival hour in the destination's local time. An hour counts only where the airport recorded at least 2,000 arrivals in it, and an airport appears only where at least twelve hours clear that bar.

ORD (Chicago) swings 15.0 min across its day, from 9.60 min at 05:00 to 24.58 min at 20:00 — the worst hour is 2.6 times the best. MCI (Kansas City) varies by 0.8 min across its whole day. The airports at the top of that table are overwhelmingly connecting hubs, which is what you would expect if the mechanism is the bank: a hub deliberately lands a large number of aircraft inside a short window so that passengers can connect between them, and there are not enough gates and taxiways to absorb them all at once. The airports at the bottom take their arrivals in a steady trickle and never have that problem. This is an association between airport type and hourly variation, not a measurement of the cause.

Carriers with at least 500 arrivals at the airport, at airports where at least four carriers clear that bar. Ranked by the gap between the slowest and the quickest.

The largest single influence on taxi-in in this data, after the airport itself, is which carrier is flying you into it. At ATL (Atlanta) an arriving Frontier Airlines flight taxis 21.96 min and an arriving Southwest Airlines flight taxis 7.40 min: same airfield, same runways, same air traffic control, 14.6 min apart. The gap is not an artefact of when each carrier arrives. Reweighting the arrivals of Southwest Airlines at ATL onto the arrival-hour distribution of Frontier Airlines, over the hours both operate, moves it only from 7.40 min to 7.23 min, against 21.96 min for Frontier Airlines on the same basis — the hour of day explains almost none of it. Nor is it one strange airport. Across the 63 airports where at least four carriers clear the comparison threshold, Frontier Airlines is the slowest carrier at 37 of them and Southwest Airlines is the quickest at 36. The obvious explanation is geometry — gates are assigned, not chosen, and a carrier holding a handful of stands at the far end of a remote concourse taxis past everyone else's to reach them, every single arrival, for years. The DOT record does not contain gate assignments, so that explanation is an inference and this report will not pretend otherwise. Aircraft type, arrival runway and the direction each carrier arrives from are also uncontrolled. What is measured is the gap itself, and the gap is large, consistent across airports, and stable across the whole period.

Taxi-out is averaged over every departure from the airport and taxi-in over every arrival into it — two different measures of the same place, not two versions of one. Rank 1 is the slowest. A positive difference means the airport is relatively slower to arrive at than to leave.

Leaving costs far more than arriving everywhere: taxi-out across these airports averages 18.67 min, departures-weighted, against 9.02 min of arrivals-weighted taxi-in. That asymmetry has a well-understood cause — a departure has to queue for a runway and an arrival does not. What is less obvious is that the two are only loosely tied together. Across the 64 airports with both measures the correlation is 0.73, so knowing how slow an airport is to leave leaves roughly 47% of how slow it is to arrive at unexplained. BUR is the 25th slowest of the 64 to depart from and only the 63rd slowest to arrive at; AUS is the reverse, 42nd to leave and 17th to arrive at. They are different constraints. Taxi-out is mostly a runway queue, which is a departure problem; taxi-in is mostly the distance from the runway to a gate that is free, which is a gate problem. An airport can be badly short of one and comfortable on the other, and several are.

Taxi-in has got worse over the period, but almost all of the deterioration is in one place, and reporting the national figure without saying so would be misleading. Arrivals-weighted across covered airports, taxi-in averaged 8.67 min in the first twelve months of the period and 9.41 min in the last — 0.74 min worse. ORD (Chicago) alone went from 14.46 min to 20.52 min, a deterioration of 6.06 min, while its arrivals rose from 262,429 to 345,796. Take ORD out and the rest of the country moves from 8.38 min to 8.68 min, a change of 0.30 min — real, but a fraction of the headline. The growth in arrivals at ORD is the obvious candidate explanation and it is not the only one; construction, runway configuration and gate availability all change over three years and none of them are in this dataset.

One appealing theory did not survive contact with the data, and it is worth showing why, because it is exactly the shape of claim this report could have made carelessly. The idea is that landing early costs you: your gate is still occupied by the aircraft that has not left yet, so you wait on a taxiway. The raw comparison looks like strong support. Arrivals whose flight left the origin gate ten or more minutes early taxied 10.35 min against 8.41 min for flights that left between on time and fourteen minutes late — nearly two minutes, across 1,450,181 flights. But early departures are not spread evenly across airports, carriers and hours. Comparing the two groups only within the same airport, the same carrier and the same arrival hour, and reweighting so the two have an identical mix, that gap all but disappears: 11.00 min against 10.58 min across 1,274 comparable groups, under half a minute. Most of the apparent effect was composition. Something small survives and it points the same way, but it is far too small to support the story, so the story is not told here. The same caution applies throughout: taxi-in as reported bundles the distance from runway to stand, waiting for a gate, waiting for a marshaller or a tug, and crossing an active runway, and the DOT record does not separate them.

Taxi-in sits inside the schedule, so a long one is a cost in time rather than proof that a flight will be late — but it is a cost charged at the worst possible moment, after the flying is done and while a connection is running out. Three things follow. If you are connecting through an airport near the top of the first table, the connection you were sold is shorter than it looks, because the walking cannot start until the taxi ends. If you have a choice of arrival time into one of the hubs in the hourly table, the middle of the day is not much better on average and is markedly less likely to hand you the twenty minute taxi. And when two itineraries into the same airport differ only by carrier, the carrier column above is a real, repeatable difference between them that no booking site will show you. None of this predicts an individual flight. It describes what has happened to 17,070,588 of them.