Rankings / Methodology

Methodology

Everything here is calculated from airline filings to the US Department of Transportation. This page lists the formulas, the choices we made and the limits.

The main figure

Passenger-miles per gallon is the number of passenger-miles an airline flew in scheduled service, divided by the gallons of jet fuel it burned in scheduled service. A passenger-mile is one paying passenger carried one mile. We use it because it counts both how big the aircraft are and how full they fly.

Two related figures sit beside it. Seat-miles per gallon uses seats offered instead of passengers carried, so it ignores how full the flights were. Load factor is passenger-miles divided by seat-miles.

CO2 per passenger-mile is gallons burned times 9.75 kg of CO2 per gallon, divided by passenger-miles. The 9.75 kg factor is the US EPA's figure for kerosene-type jet fuel in its GHG Emission Factors Hub. We apply the same factor to every year so the trend reflects fuel use and not changes in the factor.

Where the data comes from

• Fuel: BTS Form 41 Schedule P-12(a), monthly gallons of fuel consumed by carrier. We use the scheduled service total, which covers domestic and international flying.
• Traffic: BTS Form 41 Schedule T-100 segment data for US carriers. For each carrier, route and month it gives departures, seats and passengers. We keep scheduled passenger service and skip charter and cargo-only flying.
• Seat-miles are seats times the segment distance, added up over all segments. Passenger-miles are passengers times the segment distance.
• Years: only calendar years where all twelve months are present in both files. 2000 to 2025 are in the data. A part-year would flatter or penalise an airline depending on the season.
• Size cutoff: airlines with fewer than one billion seat-miles in a year are left out of that year. The airlines ranked in 2025 cover 100.0% of US scheduled seat-miles.

Regional airlines

This is the hardest part, and it matters most for United, Delta, American and Alaska. Many regional flights wear a major airline's name but are flown by a separate company under contract. SkyWest, Republic, Envoy, Endeavor and PSA are examples. In recent years those regionals report little or no fuel in their filings, because the airline that sells the seats reports it. In the traffic filings the opposite is true: the regional reports the seats and the passengers.

If you divide a major airline's fuel by its own traffic alone, the result is too low, because its fuel includes regional flying that its traffic leaves out. Earlier in the data many regionals reported their own fuel, and then the major's figure leaves the regional out entirely. Either way the answer changes when a contract changes, so the series would not be comparable over time.

We settle it the same way in every year. A fixed list of airlines that sell their own seats is published on its own. Every other scheduled passenger operator is treated as a regional, and its flying and whatever fuel it reported are moved to the airlines it flies for. Subsidiaries count as the airline itself: Horizon Air at Alaska, and Continental Micronesia at Continental.

The split works from hubs. For each airline that has regional partners, we count its own mainline seats at every airport. A regional flight between two airports is credited to the airline whose mainline presence best fits both ends. A SkyWest flight between Salt Lake City and a small town goes mostly to Delta, because Delta dominates Salt Lake City. A flight touching the hubs of two airlines is split by how well each fits.

Each regional also has an overall mix of partners. We estimate that mix by starting from an even split and repeatedly re-weighting it toward the airlines whose hubs the regional actually serves, until it settles. Regionals that fly for one airline come out at or near 100% for that airline. The table above shows the result.

How far to trust it

We can test the method against regionals whose partner is public record. 7 regional and period combinations are checked, such as Envoy and Piedmont for American and Endeavor for Delta, covering 89 regional-years. The method puts 96.2% of that flying with the right airline. The worst single case is PSA Airlines in 2016, where 73% went to American Airlines. A test fails the build if this accuracy falls.

Regionals that fly for several airlines, such as SkyWest and Republic, cannot be tested this way, and they are where the error is largest. To show it, we run the split twice: once with the partner-mix step and once without it. Airlines with regional partners get a range on the rankings, drawn as an amber line. When two airlines' ranges overlap, treat them as tied. Airlines with no regional partners, such as Southwest, JetBlue, Spirit and Frontier, have no range because nothing is estimated.

What the figures leave out

• Cargo in the hold. Fuel burned to carry freight and mail in passenger aircraft is counted against passengers. Airlines with big widebody fleets carry more belly cargo, so their passenger figure looks slightly worse than it would on a split basis.
• Other warming effects. The CO2 figure counts carbon dioxide from burning fuel. Contrails, nitrogen oxides and the emissions from making and moving the fuel are not included.
• Foreign airlines. Only US carriers file these forms. Foreign airlines flying to the US are not in the data.
• Mergers. BTS combines merged airlines' filings from a set date. A merged airline's line can jump in that year because it starts to include the other airline. We keep the earlier airline as its own entry before that date.
• Years we dropped. ATA Airlines in 2008 (105.5 passenger-miles per gallon) was left out. No jet airline gets close to that figure, which means the fuel and traffic filings do not describe the same flying, as happens when an airline stops mid-year.
• Different routes. An airline flying long routes with few stops and one flying short hops are not directly comparable. Average trip length is listed next to every ranking so the difference is visible.

Rebuilding it

All inputs are public BTS files and one EPA table. The data page has the output as a CSV. Figures update when BTS publishes a new year.