The Red-Eye Recovery

Summary

On-time arrivals fall almost without a break from 92.2% at 05:00 to 69.0% at 20:00 — the same accumulating delay behind nearly every hour of the American flying day. Then the pattern reverses. Flights scheduled from 23:00 through 04:59, the red-eye window, arrive on time 79.7% of the time, 10.1 points ahead of the 69.6% managed by the 17:00–20:59 evening peak just before them, with an average arrival delay of 5.1 min against 15.5 min. The recovery is not instant and it is not universal: the two hours right after the evening peak barely show it, and at 7 of the 30 airports with enough red-eye departures to measure, flying late at night is the worst choice of the day, not the best.

An aircraft that falls behind in the morning stays behind for the rest of its day, and the airspace and the airports fill up as the hours pass — that is why on-time performance slides almost every hour from breakfast to the evening rush. What is not obvious is that the slide has an edge. Late enough at night, the schedule stops looking like the tail end of a bad day and starts looking like a different, better one. Flights scheduled well after the evening rush are not flown by aircraft that missed the day's problems — most of them have been flying all day too. What has changed by then is the system around them: fewer aircraft competing for the same runways and the same slice of controlled airspace, and no ground-delay program metering arrivals into a hub that filled up hours earlier.

All three windows are departures measures at the origin airport, on the same basis as the "On-time arrivals" figure on an airport's own profile page. Cancellation rate is of scheduled departures in the window; every other column describes flights that operated.

The cause codes carriers file with the DOT tell the same story from a different angle. In the evening peak, 18.5% of coded cancellations are attributed to the wider air traffic system — the code for congestion, volume and airspace flow restrictions rather than anything specific to one aircraft. Overnight that falls to 5.4%, a collapse consistent with airspace that is simply not busy enough to need managing. What does not fall is the carrier's own share: it rises from 19.9% in the evening to 41.0% overnight. A cancellation that happens at 2 AM is disproportionately a maintenance or crew problem, because the one thing that reliably went away by then is other traffic.

The cancellation reason is the single code the operating carrier files with the DOT. Each window shown recorded at least 500 coded cancellations, the same floor every cause-split section on the site uses; security-coded cancellations are omitted as a column because they are too rare in any window to show a rate worth reading.

Limited to airports scheduling at least 120 departures in both the evening peak and the red-eye window. Ranked from the largest overnight improvement to the largest overnight decline.

This dataset has no aircraft tail number and no gate or crew schedule, so the exact mechanism cannot be traced the way the cause-code shift can be. The most plausible explanation is the one the data points toward — a genuinely quieter system, not a fresher aircraft — but it cannot be confirmed flight by flight. Nor does the recovery reach everywhere: at the airports with the sharpest declines, a late-night departure is worse than an evening one, which argues against any single nationwide cause and for something closer to how each individual airport uses its own late hours. And a red-eye carries a risk this table does not price in. It is very often the last scheduled departure of the day on that route, so a cancellation has no later flight the same day to fall back on — the same reliability that makes the window attractive also makes its failures more expensive.

Where the table above shows a real gap, a genuine red-eye — not a merely late evening flight, but one scheduled close to or after 23:00 — is one of the more reliable ways to fly, and the tiredness it costs is a real, known price for a real, measured benefit. Where the table shows the opposite, treat the last flight out as a gamble rather than a hack: it is booked for convenience, not for punctuality, and the data at that airport says so plainly. Either way, an itinerary with no later flight that day to recover onto should weigh the cancellation risk as heavily as the on-time rate.