Improving Fairness in Flight Delays

Airlines and the FAA don't like flight delays any more than passengers, but what's to be done? Assistant Professor Douglas Fearing and colleagues propose a "fairness" system that could save travelers time and service providers millions of dollars annually. Key concepts include:
  • The model could reduce flight delays by 4 percent or more on some of the worst travel days, resulting in a systemwide savings of $25 to $50 million annually.
by Julia Hanna

People who fly the friendly skies with any regularity have no doubt experienced the not-so-friendly feelings that come with missed connections, canceled flights, and general delays. Angry, powerless, they join the rebooking queue or camp out in the boarding area to await the latest update.

Assistant Professor Douglas Fearing doesn't have the magic bullet to make this painful part of air travel a thing of the past. But his research on evaluating and improving the performance of complex systems such as air traffic control offers road warriors some comforting insights based on hard analysis that help demystify the frustrating experience and could, if adopted, result in improvements to the US air traffic control system.

Fearing's interest in improving performance extends to how a large governmental organization like the Federal Aviation Administration can better work with airlines to ease congestion at airports and in the air. In Equitable and Efficient Coordination in Traffic Flow Management, a paper recently accepted for publication in Transportation Science and coauthored with Cynthia Barnhart of MIT and Dimitris Bertsimas and Constantine Caramanis of the University of Texas at Austin, Fearing develops a "fairness" metric that (in the case of severe weather, for example) balances the trade-off between the most efficient and the most equitable allocation of delays.

The FAA's current Traffic Flow Management system, used when certain airports or flight paths become too congested, relies on the use of the Ground Delay Program (GDP), introduced in 1981, and the Airspace Flow Program (AFP), introduced in 2007. When a GDP event is implemented, such as during the recent Hurricane Irene sweep up the East Coast, the FAA coordinates ground departure times to control the arrival rate into a congested airport; in an AFP, the FAA uses the same approach to adjust the amount of traffic flying into congested airspace. These two tools account for approximately 30 percent of all air transportation delays, according to the researchers.

“If you manage this network of disruptions using an optimization-based approach, you can get more efficiency out of the system and still have a nice balance of fairness”

In the event of significant delays in the system, the FAA allocates arrival slots to airlines based on the agency's definition of fairness—priority is on a first scheduled, first served basis. Airlines can then respond by making changes within their allocated slots before the final schedule is approved. Because GDPs and AFPs are implemented separately, however, a flight affected by a congested airport and a congested airspace may receive a conflicting time of departure. Typically, the GDP is given priority over the AFP, although a relatively recent development in flight scheduling allows traffic flow managers to exempt AFP-directed flights from subsequent GDPs. The problem: for the sake of efficiency, resulting schedules sometimes deviate from the first scheduled, first served process.

"This approach ignores the fact that there is a networked structure where a flight may go through a congested airspace before landing at an airport," Fearing says. "What we show is that if you manage this network of disruptions using an optimization-based approach, you can get more efficiency out of the system and still have a nice balance of fairness when you have two conflicting programs in the system."

According to the US Congress Joint Economic Committee, the cost associated with domestic flight delays in the United States during 2007 was estimated at $25.7 billion ($12.2 billion in increased airline operating costs, $7.4 billion in passenger time lost, and $6.1 billion in costs to related industries). On about 40 percent of the days in 2007, there was at least one GDP and at least one AFP in place (only 16 days were miraculously free of either program).

Fearing estimates that the model he and his coauthors developed—called the Ration-by-Schedule Exponential Penalty model—could reduce flight delays by 4 percent or more on some of the worst travel days, resulting in a systemwide savings of $25 to $50 million annually.

"Even small improvements result in pretty substantial savings," he observes. "It's not a bigger number because we're only getting those improvements when there are conflicts between the GDP and AFP programs. But making the worst days better would still be a nice improvement on the system."

For fliers, the RBS-EP proposal would help those whose flights would have been given unfair assignments based on the current approach: based upon a new metric for defining fairness, they would receive earlier (and therefore more advantageous) departure slots.

The Best And The Worst

In a second working paper, "Modeling Passenger Travel and Delays in the National Air Transportation System," Fearing and coauthors Cynthia Barnhart and Vikrant Vaze of MIT apply statistical techniques to develop a representative sample of individual passenger flows through the system in 2007 and then extend a methodology developed by Bratu and Barnhart to estimate the corresponding passenger delays.

Some of their findings confirm what veteran travelers have often suspected, while others offer surprising insights. Some highlights and lowlights:

Regional airlines the worst. Flight delays were highest for regional airlines such as American Eagle, Skywest, and Mesa Airlines, which also have the greatest percentage of cancelations and connecting passengers.

Low-cost airlines soar. Delay rates were lowest for low-cost carriers such as Southwest, JetBlue, AirTran, and Frontier, since these airlines tend to have fewer connecting passengers and often fly into airports that are less capacity constrained. Legacy carriers fell somewhere in-between.

Worst airports for connections. Passengers scheduled to transfer in Newark, Chicago O'Hare, New York's La Guardia and Kennedy, Washington Dulles, or Philadelphia experienced the longest average connecting passenger delays, with the itineraries of over 10 percent of passengers disrupted. The worst transfer airports based on departure cancelation rates included Washington's Reagan, Boston's Logan, and Dallas/Fort Worth International airports. At Hartsfield-Jackson Atlanta, Chicago O'Hare, and Dallas/Fort Worth airports—the airports with the highest number of planned passenger connections—40 percent of all domestic passengers' missed connections and 43 percent of all disruptions to one-stop passengers occurred.

Don't sleep in. If you don't like getting up early to catch a flight, consider that the average delay for passengers departing after noon is 86.8 percent greater than for those flying in the morning (obviously, passengers disrupted in the morning also have more re-booking alternatives). In a related finding, Monday and Saturday have the fewest passenger delays, due in part to the higher percentage of morning passengers on these two days.

Delays higher in summer and winter. Delays are worst in the summer (June, July, August) and winter (December, January, February) months, with passenger delays 56 percent higher at these times than in the remaining six months. The average passenger delay in the summer months was 37.4 minutes; in winter, 36 minutes; for the remaining six months, 23.5 minutes. September and November were the only two months with average passenger delays of less than 20 minutes.

The more flights the better. Passengers benefit by flying airlines offering more flights to their destinations. Fearing and his coauthors find that the average delay to disrupted non-stop passengers on routes with at least 10 daily flights per carrier is 31.4 percent lower than the overall average for disrupted passengers. The average delay on routes with at most three daily flights per carrier is 15.3 percent higher than the overall average.

The Best On-time Airline

If avoiding delays is important to your travel plans, Southwest may be the way to go. Long recognized for its on-time performance, Fearing and his coauthors show that the airline's average passenger delay is nearly 55 percent lower than its competitors (15.6 minutes vs. 33.7 minutes), although its average flight delay is only 36.3 percent lower than other airlines. Only 1 percent of Southwest's flights are canceled as compared to 2.8 percent for other carriers.

"Southwest also flies into airports that are less capacity constrained, they have fewer connecting passengers, and their schedules tend to be flatter over the day," Fearing observes. "Many of the legacy carriers will have banks of arrivals followed by departures, which creates a larger network of possible connecting options. Southwest is more focused on direct travel. Their schedules lead to longer connection times with less opportunity for a missed flight."

Fearing says that future work will consider schedule disruptions from the perspective of the airlines' decision-making process. "When you have severe disruptions, how do you decide whether to cancel or delay a flight—and which flights do you cancel?" he explains. "I like using statistics and computational methodologies to solve big, hard problems, but to do so in a way that also generates some interesting insights."

In another line of research, Fearing is working with MIT faculty member Steve Graves, Ph.D. student Jason Acimovic, and Columbia University professor Mark Broadie to develop "Strokes Gained Putting," a new metric for the Professional Golfers' Association that corrects for distance and difficulty biases to measure how much a golfer benefits from his or her putting game.

About the Author

Julia Hanna is Associate Editor of the HBS Alumni Bulletin.
    • Roger McDonald
    • retired
    I spent 26 years in Air Traffic Control and 7 years as a
    pilot in the system.. ALL delays are caused by runway
    acceptance rate and ground operations and parking
    space.. ALL delays.. The Air Traffic Center could drown
    ANY airport with aircraft..
    • Kenneth Geisinger
    • FAA retired
    I would very much like to obtain an electronic copy of these papers. I worked in these areas for several years while at the FAA.
    • Pardha Saradhi
    Given assumptions - Limited resources, Limited parking space, Limited time to provide the runway for incoming aircraft with zero risk of unwanted accidents or collisions.

    My suggestions - Will be happy if it finds its place in case study at HBS class or throw open to debate

    1. Clear demarcation of incoming airlines and outgoing ones - Airport to be designed / modified accordingly

    2. Distance to be minimum half a kilometer between incoming land point and take off point.

    3. Logistics movement to be automatized to reduce the time delays - Cargo or passenger belongings.

    4. Hop flights to be sufficiently refuelled to avoid delays.

    5. Shuttle flights to be swapped and not returned. - To given an example, if a flight to New York from California leaves at 9:00 AM to reach California by 12:00 PM then return flights should be a different one ready to leave California back to New York. This means two flights operating instead of one. Except the fixed cost maintenance the other costs would remain same as crew would be same

    5. We cannot control the weather - but we can definitely work around for avoiding the delays in areas controlled by operations. This would help the throughput for better air traffic.

    6. Mathematical modelling - traffic parameters, logistics parameters, time efficiency parameters & lean timeperiod parameters - These 4 parameters shold be modelled with 1, 2 & 4 being minimised and 3rd parameter to be maximised.
    • Paul Robichaux
    • seated in coach somewhere
    A cursory examination of the items listed in the section titled "The best and the worst" revealed zero surprises to anyone who's done any volume of business travel. A quick review of basic weather knowledge will tell you that summer delay patterns correspond to the times and locations where convective weather's most common-- afternoons in the summer. The parallels to winter weather is left as an exercise for the reader.
    • Kevin Johnson
    • left in limbo
    It may be useful to look at peak passenger delays as well. Returning from the Caribean through Charlotte on a major carrier a number of passengers including my family were bumped from a connecting flight due to a system failure when we all checked in.

    Once bumped, we were told that all flights to our destination on all airlines were booked for three days and that rather than bumping anyone else, we'd have to wait up to three days for the next open seats. A couple of us ended up renting cars and driving to DC to catch a flight.

    I wonder how this type of delay is reflected in your passenger delay statistics.