Skip to the content

Disruptions

Did the detector see the meltdowns coming, and how fast did each carrier recover?

The detector scores every carrier's day, and every day at the 30 busiest airports, against its own previous four weeks. Its threshold was chosen on 2015 to 2022 by the cost of a false alarm against a missed disruption, and graded on the reporting years against a table of known disruptions, each with a citation.

The known disruptions and when they were flagged

EventOnsetWindowUnitsFlaggedFirst alertDays after onset
Boeing 737 MAX grounding2019-03-13fittingWN;AA;UAyes2019-03-130
COVID-19 collapse in air travel2020-03-15fittingallyes2020-03-172
Winter Storm Uri in Texas2021-02-14fittingDFW;DAL;IAH;HOU;AUS;SATyes2021-02-140
Southwest scheduling crisis2022-12-21fittingWNyes2022-12-221
Winter Storm Elliott2022-12-21fittingallyes2022-12-221
FAA NOTAM system outage2023-01-11reportingallyes2023-01-110
Alaska 1282 and the 737-9 grounding2024-01-06reportingAS;UAyes2024-01-060
CrowdStrike outage and Delta's recovery2024-07-19reportingDLyes2024-07-190
Hurricane Milton2024-10-08reportingTPA;MCO;SRQ;RSW;PIE;SFByes2024-10-080
Newark radar outages and flight caps2025-04-28reportingEWRyes2025-05-013
Alaska Airlines IT outage and ground stop2025-10-23reportingASyes2025-10-230
FAA flight reductions during the federal shutdown2025-11-07reportingallyes2025-11-081
January 2026 winter storm2026-01-24reportingallyes2026-01-240

Source: BTS Reporting Carrier On-Time Performance (real); model: detector; population: every scheduled flight, 2015 to 2026-07; as of 2026-10-01.

The operating point

A false alarm costs 1.0 and a missed disruption 50.0; the threshold with the least total cost on the fitting years was 29.0, inside the grid. On the reporting years it flagged 8 of 8 disruptions with 297 false alarm episodes.

The cost curve bottoms out at a threshold of 29.0

Total cost on the fitting years by alert threshold: false alarm episodes plus missed disruptions at their stated costs

Cost

02,0004,0006,0008,00010,0005.010.015.020.025.030.0Alert threshold, robust standard deviationsTotal cost

Source: BTS Reporting Carrier On-Time Performance (real); model: detector; population: every scheduled flight, 2015 to 2026-07; as of 2026-10-01.

The two carrier meltdowns

Southwest Airlines took 9 days to get back to its peers and Delta Air Lines 5; across every alert episode, how fast a carrier recovers is a trait of the carrier

Daily cancellation rate of the carrier against other carriers at its ten busiest airports

Southwest Airlines, December 2022

Share of scheduled flights cancelled

the disruption0%20%40%60%80%2022-12-022022-12-142022-12-252023-01-06Dayback to its peers after 9 daysSouthwest AirlinesOther carriers at the same airports
  • Southwest Airlines
  • Other carriers at the same airports

Delta Air Lines, July 2024

Share of scheduled flights cancelled

the disruption0%10%20%30%40%50%2024-06-212024-07-032024-07-142024-07-262024-08-07Dayback to its peers after 5 daysOther carriers at the same airportsDelta Air Lines
  • Delta Air Lines
  • Other carriers at the same airports
Threshold 29.0 chosen on 2015 to 2022 (inside the grid); 8 of 8 reporting year events flagged.

Source: BTS Reporting Carrier On-Time Performance, model: the disruption detector, every scheduled flight, 2015 to 2026-07, as of 2026-10-01.

The alerts in the reporting years

UnitFirst dayDaysPeak score
airport TPA2024-09-262166.7
airport TPA2024-10-084166.7
airport MCO2024-10-093166.3
airport DCA2026-01-254166.1
airport IAH2025-01-212166.1
airport JFK2026-02-223166.1
airport DAL2023-01-304164.2
airport BWI2026-01-252163.2
airport DFW2026-01-245158.6
airport IAH2024-07-082157.5
airport PHL2026-01-254155.7
carrier OH2026-01-245154.4
carrier YX2026-01-244151.1
airport EWR2026-01-252150.4
airport CLT2026-01-245149.6
airport JFK2026-01-253146.9
airport DCA2025-01-062146.7
airport CLT2026-01-312146.0
airport DFW2025-01-092142.1
airport BNA2026-01-244141.5
airport DFW2023-01-304140.2
airport ATL2025-01-102139.9
carrier YX2026-02-223133.7
carrier B62026-01-253130.6
airport BOS2026-01-253130.2

Source: BTS Reporting Carrier On-Time Performance (real); model: detector; population: every scheduled flight, 2015 to 2026-07; as of 2026-10-01.

Recovery by carrier

CarrierAlert episodesMedian daysMean daysLongest
US Airways11.0 days1.0 days1
Virgin America41.0 days1.2 days2
Endeavor Air341.0 days1.7 days10
Republic Airways421.0 days1.9 days11
Envoy Air251.0 days2.0 days10
Alaska Airlines91.0 days2.4 days10
ExpressJet Airlines101.0 days2.5 days11
American Airlines141.5 days2.6 days9
JetBlue Airways342.0 days2.0 days6
PSA Airlines232.0 days2.4 days11
Delta Air Lines202.0 days2.5 days12
United Airlines202.0 days2.7 days13
Frontier Airlines132.0 days2.8 days14
Southwest Airlines92.0 days2.9 days8
Allegiant Air132.0 days3.2 days14
SkyWest Airlines42.0 days3.8 days10
Spirit Airlines162.5 days2.8 days7
Horizon Air23.0 days3.0 days3
Mesa Airlines44.0 days5.2 days12
Hawaiian Airlines65.0 days6.5 days14

Source: BTS Reporting Carrier On-Time Performance (real); model: detector; population: every scheduled flight, 2015 to 2026-07; as of 2026-10-01.

What the ops director would push back on

The known events table is short, so a disruption that is real but not in the table counts as a false alarm, which overstates the false alarm rate. The peer control in the event studies is the other carriers at the same airports, who were hit by the same storm and carried some of the stranded carrier's passengers; the storm is differenced out, and the published excess is a floor.