American blamed weather for 94% of its May cancellations; Delta blamed itself for 91% of its own — two carriers the same size, cancelling at nearly the same rate, filing opposite reasons

The U.S. Department of Transportation's on-time database for May 2026 — the most recent month posted (file dated 2026-06-30) — records 611,735 scheduled domestic flights by the reporting carriers: 5,655 cancelled (0.92%), 1,734 diverted, and 604,346 completed, of which 78.6% arrived within 15 minutes of schedule. Every cancelled flight carries a one-letter cause code the airline itself assigns: A (air carrier), B (weather), C (National Air System), D (security). Across all 5,655 cancellations the split was 3,319 weather (59%), 1,907 carrier (34%), 428 NAS (8%), and 1 security. Two carriers of nearly identical size coded their cancellations in near-opposite ways. American Airlines flew 88,740 scheduled flights and cancelled 1,324 (1.49%); 1,245 of those — 94.0% — were coded weather, and just 64 — 4.8% — carrier. Delta flew 91,955 and cancelled 1,144 (1.24%); 1,044 of those — 91.3% — were coded carrier, and 73 — 6.4% — weather. Same month, comparable fleets, similar cancellation rates, mirror-image reasons on file. American's weather cancellations are concentrated, not spread. Of its 1,324 cancellations, 1,082 (82%) fell on just five days — May 10, 19, 20, 24 and 25 — when its systemwide cancellation rate was 7.57%, against 0.33% on the other 26 days. And 531 of its 1,324 cancellations (40%) originated at a single airport, Dallas/Fort Worth, American's largest hub (16.6% of its departures). Nationwide, American's 1,245 weather cancellations were 37.5% of every weather-coded cancellation in the country that month; add its two wholly-owned regional carriers, Envoy (437) and PSA (368), and the American system alone accounted for 2,050 of the 3,319 national weather cancellations — 61.8%. On the five storm days, the DFW record shows the cause code was unanimous and the cancellation rate was not. American cancelled 504 of its 2,367 DFW departures (21.3%), coding 503 of them weather. Its regionals cancelled at similar rates and also coded weather: Envoy 178 of 935 (19.0%), PSA 72 of 292 (24.7%). Independent regional SkyWest cancelled 73 of 294 (24.8%). The other majors flying DFW those same days cancelled far less, and also called it weather: Delta 10 of 158 (6.3%), Frontier 10 of 171 (5.8%), United 4 of 148 (2.7%). Every carrier at the airport blamed the sky; the share of each carrier's schedule that fell ranged from 3% to 25%. Across the full month at DFW, American cancelled 3.61% of its 14,714 departures, Delta 1.74% of 979, United 1.00% of 901.

The following is analysis, not fact. The headline divergence — 94% weather for American, 91% carrier for Delta — is a real difference in the record, but it is not evidence that either carrier mislabeled anything ^mmodeled. Reading the flight-level data resolves it: American's cancellations are overwhelmingly a hub-storm phenomenon. Four of every five fell on five days, two of every five originated at Dallas/Fort Worth, and on those days every carrier at DFW — Delta and United included — coded the same weather cause. The unanimity across seven operators is the strongest available signal that the weather was genuine ^mmodeled. American's high weather share, then, is largely a statement about *where and when* it cancels, not about how it labels a given event. What differs under the identical label is exposure. On the storm days American scrubbed 21% of its DFW schedule while Delta, United and Frontier cancelled 3 to 6% at the same airport. The mechanism is structural: a hub carrier banks hundreds of tightly-connected flights through one airport within a few hours, so a convective cell parked over that airport during a bank cascades into mass cancellations; a carrier running a handful of point-to-point flights through the same field has far less exposed to the same storm ^mmodeled. That American operates the largest bank at DFW, and its two wholly-owned regionals feed it, is why the American system produced 62% of the nation's weather cancellations from one month's weather ^mmodeled. "Weather cancellation" is as much a description of an airline's network design as of the atmosphere. The label is not cosmetic. DOT's Airline Customer Service Dashboard sorts cancellations into "controllable" — within the carrier's control, such as maintenance or crew — and everything else; weather and air-traffic causes fall outside it, and the amenities a carrier commits to (rebooking, meals, hotels) attach to the controllable ones. By that yardstick American recorded about 5% of its May cancellations as controllable and Delta about 91%. The naive suspicion runs one way — a carrier has every incentive to code weather and dodge the obligation — but the data cuts against reading American's number that way: its weather cancellations sit exactly where and when a real storm would put them, and Delta, the carrier coding *more* of its cancellations as its own fault, is the outlier against incentive, not American ^mmodeled. The honest read is narrower: the reason on your cancellation notice is set partly by the weather and partly by whether your airline chose to route you through a hub bank in storm season.

The strongest counter is that this compares operations that are not alike. American runs a connecting-bank hub at DFW; Delta, United and Frontier fly it as a spoke with a few point-to-point departures a day. A hub bank and a spoke are differently vulnerable to the same storm by construction, so the 21%-versus-6% gap is partly a comparison of network types, not of how well two airlines flew the same weather. The per-carrier DFW samples for the non-American majors are also thin — Delta 158, United 148, Frontier 171 departures across the five days — so their storm-day rates swing on a handful of flights. Two more cautions. The cause code is self-assigned and unaudited; the cross-carrier agreement at DFW is good evidence the weather was real, but it cannot rule out that some flights coded weather had a controllable trigger (a crew or aircraft already out of position when the storm arrived), nor that a flight coded carrier elsewhere was in part weather-driven — the record does not show the underlying cause, only the letter filed. And one month is one month: May is convective season in north Texas, so a carrier hubbed at DFW will look weather-heavy in May and might look different in January. The durable, well-sampled claims are the aggregates — 94% versus 91% coding on nearly identical fleets, 82% of American's cancellations on five days, 40% at one airport, and 62% of the nation's weather cancellations from the American system — not any single carrier's storm-day rate at one airport.

A traveler reads "cancelled — weather" and hears an act of God, something no airline could help. The record complicates that. On the same afternoon at the same airport under the same storm, the fraction of a carrier's flights that survived ran from three-quarters to nineteen in twenty, and the difference tracked how many flights each airline had chosen to stack through that hub in that hour. The weather is not negotiable; the exposure to it is a schedule, written months in advance. DOT's dashboard draws its line at "controllable," and weather sits on the far side of it — but the number of flights a storm can cancel is decided by a planner, not a cloud.

A single cancellation notice tells you the reason and nothing else. Reading all 611,735 May flights at once shows the reason is a composite: American filed "weather" on 94% of its cancellations and Delta filed "my fault" on 91% of its own, two airlines the same size cancelling at nearly the same rate — and the gap dissolves the moment you see that 82% of American's cancellations fell on five days and 40% at one hub, where every carrier, Delta included, also blamed the weather. What the whole record reveals that no boarding pass can is the part the cause code hides: under one storm at one airport, the airlines cancelled anywhere from 3% to 25% of their flights, and the number that decided which travelers got stranded was not the weather they shared but the count of flights each had banked through that hub — a scheduling decision wearing a meteorologist's label.