Airline Mitigation of Propagated Delays via Schedule Buffers:

Theory and Empirics

by

Jan K. Brueckner
University of California, Irvine

Achim I. Czerny
Hong Kong Polytechnic University

Alberto A. Gaggero
University of Pavia

September 2019, revised January, December 2020

Abstract

This paper presents an extensive theoretical and empirical analysis of the choice of schedule
buffers by airlines. With airline delays a continuing problem around the world, such an under-
taking is valuable, and its lessons extend to other passenger transportation sectors. One useful
lesson from the theoretical analysis of a two-flight model is that the mitigation of delay prop-
agation is done entirely by the ground buffer and the second flight’s buffer. The first flight’s
buffer plays no role because the ground buffer is a perfect, while nondistorting, substitute. In
addition, the apportionment of mitigation responsibility between the ground buffer and the
flight buffer of flight two is shown to depend on the relationship between the costs of ground-
and flight-buffer time. The empirical results show the connection between buffer magnitudes
and a host of explanatory variables, including the variability of flight times, which simulations
of the model identify as an important determining factor.
 Airline Mitigation of Propagated Delays via Schedule Buffers:
Theory and Empirics
by

Jan K. Brueckner, Achim I. Czerny and Alberto A. Gaggero∗

1. Introduction
Flight delays are a worldwide problem, a consequence of the substantial growth in air
travel over recent decades. In the US alone, the cost of delays for passengers and airlines was
estimated at $32.9 billion in 2010 by Ball et al. (2010). In response to the problem, the US
Department of Transportation requires all major US airlines to provide monthly information
about delays, which generates widely viewed on-time rankings of the carriers. The European
Union has imposed rules for passenger compensation and assistance in the event of long flight
delays.
A major source of flight delays is airport congestion, which is the subject of a large literature
(see Zhang and Czerny (2012) for a survey). But whether congestion leads to flight delays is
largely under the control of the airlines, since they are free to set scheduled flight durations.
In other words, the congestion-related lengthening of flight times can be built into airline
schedules through a practice known as “schedule padding,” whose recent growth is documented
by Forbes, Lederman and Yuan (2019) and others. While airport congestion may make flights
longer, this schedule adjustment prevents them from arriving late.
Despite this overall adjustment in response to broad trends, flight times are still influenced
by many random daily factors, including weather, mechanical issues, and unanticipated conges-
tion, which can vary by day and hour around some expected level. Airline scheduling decisions
take account of these random influences through the choice of “schedule buffers.” One type of
buffer is known in the airline industry as a “block-time buffer” (we call it a “flight buffer”),
and it equals the amount of time added to the shortest feasible flight time to get the scheduled
arrival time. While a longer flight buffer reduces the chance of late arrival, it also makes an
∗ We thank Tiziana D’Alfonso and several referees for helpful comments and Alex Luttmann for comments
and for providing us with one of his data sets. Any shortcomings in the paper are our responsibility.

1
early arrival more likely, and while passengers dislike delays, they also do not want flights to
routinely arrive early, an outcome that leaves time gaps that must be filled. Longer scheduled
flight times also raise the airline’s (planned) operating costs, such as the cost of crew time.
In setting its flight schedules, an airline will take all three factors (disutilities from lateness
and earliness as well as operating costs) into account. Flight buffers are typically positive,
reflecting a greater concern about late as opposed to early arrivals on the part of the carrier
(following passenger preferences). Note that these same three elements also affect scheduling
by other transportation providers, such as passenger railroads and intercity bus lines.
Delays depend on more than just the operating time of a particular flight. If the incoming
aircraft arrives late, then the outbound flight is likely to depart late, possibly leading to its late
arrival even if operating time is normal. Late-arriving aircraft are in fact the major source of
flight delays, as seen in Figure 1, accounting for more delays than mechanical and crew-related
delays (“air carrier delays”) or weather.1 This type of delay is known as “propagated delay”
since it propagates from one flight to another, and it is also present in the railroad and bus
contexts.
Propagated delay can be addressed through a long flight buffer, which reduces the chance
of a late-arriving aircraft, but another tool is the “ground buffer.” This buffer is defined as
the difference between the scheduled ground time and the shortest feasible aircraft turnaround
time. A long ground buffer can absorb a late arrival of the inbound aircraft, allowing the next
flight to depart on time despite this disruption. The flight buffer for the outbound flight can
also address delay propagation, allowing the flight to arrive on time even if it departs late.
Lengthening this flight buffer is costly, however, and longer ground times are also costly since
they require more gate space.2
The purpose of the present paper is to analyze an airline’s choice of flight and ground
buffers, both theoretically and empirically. Our theoretical framework differs from most other
models because it is stylized, rather than fully realistic, and starts from first principles, treating
1 This figure shows the apportionment of total US flight delays to different causes.
2 Our analysis ignores the possibility that ground times may depend on passenger scheduling preferences,
which are not present in the model. For instance, airlines may schedule a later departure and accept longer
aircraft ground time if the result is a departure closer to the passenger’s preferred departure time.

2
propagated delay in a setting with just two flights. Thus, the airline chooses two flight buffers
and one ground buffer, taking passenger disutilities from lateness and earliness into account
along with buffer costs. The analysis yields a number of insights. A principal result is that the
flight buffer for flight 1 plays no role in mitigating delay propagation, which is instead handled
by the ground buffer and flight 2’s buffer. The reason is that, while flight 1’s buffer and the
ground buffer can both address delay propagation, the former distorts flight 1’s scheduled
arrival time, making use of the ground buffer preferable. The apportionment of the mitigation
roles between the ground buffer and flight 2’s buffer depends on buffer costs. Numerical
examples show how buffer magnitudes are affected by the variances of the random factors
affecting the operating times of the two flights, showing that flight-time greater variability
raises the flight buffers while having a complex effect on the ground buffer. Given the parallel
to other transportation sectors, the paper’s theoretical findings extend beyond the airline
industry.
The empirical work relies on US Department of Transportation data showing the daily
operations of thousands of commercial aircraft. These data allow computation of flight and
ground buffers over an aircraft’s operating day, which are then related to exogenous explanatory
factors suggested by the model. For example, one set of regressions relates the magnitude of
the flight buffer to the standard deviation of operating times for a particular flight (computed
across the months of the sample for that flight’s operations in the previous year). The empirical
work provides confirmation of some of the hypotheses suggested by the model while providing
general insight into the determinants of flight and ground buffers.
The paper is related to several strands of previous work. Earlier papers in operations
research, including Deshpande and Arikan (2012) and Arikan, Deshpande and Sohoni (2013),
present models of the choice of flight buffers, noting the similarity to the classic newsvendor
problem of Whitin (1955) (a flight’s early (late) arrival is analogous to a vendor ending up
with a surplus (shortage) of newspapers). Deshpande and Arikan (2012) consider US airlines
and use the newsvendor approach to estimate the airlines’ ratios of earliness to lateness costs.
They show that flight buffer choices depend on carrier types, route market shares, and route
characteristics. Arikan, Deshpande and Sohoni (2013) develop schedule robustness measures

3
for airline networks and use them to show how US carriers use flight and ground buffers to
absorb delay propagation. Zhang, Salant and Van Mieghem (2018) present an analysis related
to those of Deshpande and Arikan (2012) and Arikan, Deshpande and Sohoni (2013). They
show that the historical evolution of flight durations cannot explain increases in scheduled
ground and flight times in the US and conclude that these increases instead have strategic
motivations. Kafle and Zou (2016) theoretically decompose delays into propagated and newly-
formed delays, and their empirical work investigates the determinants of propagated delays.3
In the economics literature, Wang (2015) offers a different theoretical approach to the
choice of ground buffers, relating this choice to the level of competition while providing empir-
ical evidence of this link. In other empirical work by economists, Forbes, Lederman and Yuan
(2019) use a much larger dataset to confirm the earlier finding of Shumsky (1993) showing
that US carriers have added schedule padding (buffer time) over the years to improve their
on-time performance. Forbes, Lederman and Wither (2019) show that this effect is stronger
when airlines are large enough for required reporting of on-time performance, a criterion that
excludes many regional carriers.4 Complementing these schedule-padding papers, transporta-
tion engineers have also extensively studied the determinants of scheduled block times, with
contributions by Sohoni, Lee and Klabjan (2011), Hao and Hansen (2014), Kang and Hansen
(2017, 2018), and Wang et al. (2019). Other studies by economists, which are not directly
linked to our work, show the connection between market structure (mainly competition) and
on-time performance (see, for example, Mazzeo (2003) and Prince and Simon (2015)).
The plan of the paper is as follows. Section 2 analyzes choice of the flight buffer in an
introductory model with just one flight, where delay propagation is not an issue, and section
3 analyzes the two-flight model. Section 4 presents numerical examples, while section 5 offers

3 In addition, AhmedBeygi, Cohn and Lapp (2010) use a linear programming approach to show how read-
justment of ground buffers, with no increase in total ground time among a set of flights, can achieve a reduction
in propagated delays. In contrast to our stylized approach, they use a detailed and realistic model and rely
on simulation. Barnhart and Cohn (2004) highlight the importance of operations research methods for the
airline industry and describe the (at the time) state-of-the-art in airline schedule optimization. They further
highlight the importance of schedule robustness and delay propagation (which they call “snowballing”) in the
airline industry.
4 For analysis of the incentives for integration of regional and mainline carriers and its impacts, see Forbes
and Lederman (2009, 2010).

4
a model extension that incorporates connecting passengers. Section 6 describes the empirical
setup, and section 7 presents the regression results. Section 8 offers conclusions.

2. Buffer choice for a single flight

The analysis starts by considering the buffer choice for a situation where the airline operates
only a single flight, denoted flight 1. After this analysis is complete, the focus turns to the
case where two flights are operated. While delay propagation is not an issue in the single-flight
case, it is a crucial factor when two flights are operated. It should be noted that the models
we analyze are highly stylized. They are designed to expose the economic incentives faced by
airlines in the choice of schedule buffers, without being fully realistic.
Flight 1 departs at time 0 and has an uncertain duration. If no outside influences affect the
flight’s operation, its duration is given by m1 . Outside influences such as weather, mechanical
issues, and unanticipated congestion can cause the flight duration to exceed m1 , with the actual
duration equal to m1 + 1 , where 1 is a continuous, positive random variable with support
[0, 1 ]. Note that while the congestion effect is partly random, appearing in 1 , a persistently
high level of airport congestion would lead to a large value for m1 .
In scheduling the flight’s arrival time, the presence of this random term will lead the airline
to include a flight buffer, denoted b1 , which is added to m1 . The scheduled arrival time of the
flight is thus ta1 = m1 + b1 , whereas the actual arrival time, which captures the random effect,
is bta1 = m1 + 1 . The flight is late in arriving if bta1 > ta1 , or if 1 > b1 , and it is early if
1 < b1 . Passengers dislike being late or early, valuing a minute of late time by the amount x
and valuing a minute of early time by the amount y. Therefore, if the flight is late, it generates
disutility equal to x(1 − b1 ), whereas disutility is y(b1 − 1 ) if the flight is early.
The probability that flight 1 is late is equal to the probability that 1 exceeds b1 , which is
given by
Z 1
Pr(bta1 > ta1 ) = f1 (1 )d1 = 1 − F1 (b1 ), (1)
b1

where f1 is the density and F1 the cumulative distribution function of 1 . Conversely, the
probability of an early arrival is F1 (b1 ). The passenger’s expected disutility from earliness or

5
lateness is equal to

Z 1 Z b1
Ω1 = x (1 − b1 )f1 (1 )d1 + y (b1 − 1 )f1 (1 )d1 , (2)
b1 0

where the first integral captures lateness and the second captures earliness. Passenger disutility
from late and early arrivals affects fare revenue, and assuming that the airline seeks to maximize
profit, it takes into account this disutility.5
But the airline also incurs planned operating costs from scheduled flight time and ground
time. Scheduled flight costs include expenditures on fuel and crew salaries, while ground costs
consist mainly of gate rental costs. To facilitate comparison with the two-flight model, suppose
that the aircraft is only flown for part of the day, sitting on the ground for the remaining time,
so that the airline has “excess capacity” (an alternate assumption is used below in the two-
flight model). With T denoting the length of the day, ground time is T − (m1 + b1 ) > 0,
where m1 + b1 is scheduled flight time. The capital cost of the aircraft is sunk, but letting
cf denote the operating cost per minute of scheduled flight time and cg denote the cost of
ground time, total operating costs are cf (m1 + b1 ) + cg (T − (m1 + b1 )). The goal of the profit-
maximizing airline is to minimize the sum of this expression and Ω1 by choice of b1 , which
(ignoring constant terms) means minimizing Ω1 from (2) plus b1 (cf − cg ).
The first-order condition for this minimization problem is

∂Ω1
= −[1 − F1 (b1 )]x + F1 (b1 )y + cf − cg
∂b1
 
x
= (x + y) F1 (b1 ) − + cf − cg = 0, (3)
x+y

and the second-order condition, which requires (x + y)f1 (b1) > 0, is satisfied. Rearranging (3),
5 Letting G denote the benefit from travel, the net benefit from a flight, denoted V , is equal to G minus the
distutility expression in (2). Letting C denote expenditure on a nontravel good, overall passenger utility equals
U = C + V = Y − P + V, where Y is income and P is the fare for travel. The airline sets the fare at a level
that makes the consumer indifferent between traveling and not traveling, which yields utility Y (in this case
V − P = 0). The fare for a flight is then equal to V , or G minus (2), and assuming that the passenger count
is normalized to 1, fare revenue equals this expression.

6
the optimal buffer, denoted b∗1 , satisfies

x + cg − cf
F1 (b∗1 ) = , (4)
x+y

a formula similar to one derived by Deshpande and Arikan (2012). Differentiation of (4) yields

∂b∗1 ∂b∗1 ∂b∗1 ∂b∗1

> 0, < 0, > 0, < 0. (5)
∂x ∂y ∂cg ∂cf

with a greater disutility from lateness raising the buffer (thus reducing the chance of lateness)
and a greater disutility from earliness reducing it. Similarly, a higher cf (cg ) reduces (increases)
b1 .

3. The two-flight model

3.1. The setup
The airline is now assumed to operate two flights using the same aircraft, so that a delay
for flight 1 can cause lateness of flight 2. While the existence of two flights introduces the
possibility that some passengers connect from one flight to another, we assume initially that
connections are absent, showing later in the paper how they affect the analysis.
Following the single-flight assumptions, m2 denotes the undisrupted flight time for flight
2, with the actual flight time given by m2 + 2 . The positive random term 2 has density f2 ,
cumulative distribution function F2 , and support [0, 2 ]. The flight buffers are b1 and b2 , so
that the scheduled arrival times of flights 1 and 2 are ta1 = m1 + b1 and ta2 = td2 + m2 + b2 ,
where td2 is flight 2’s scheduled departure time. The scheduled aircraft ground time is denoted
tg , and the ground buffer, given by bg = tg − tg , is the excess of ground time over the minimum
feasible aircraft turnaround time, denoted tg . The size of the ground buffer is thus set by
choice of tg .6 The scheduled departure time of flight 2 is then td2 = ta1 + tg = m1 + b1 + tg ,
6 Note that, unlike flight times, ground time does not have a stochastic component, which would increase the
complexity of the analysis. Random ground time could be a consequence of weather at the airport, which could
lead to a delayed departure exactly like a late inbound flight. However, in the simpler model of Brueckner,
Czerny and Gaggero (2020), where 1 and 2 are discrete rather than continuous random variables, the effect
of random ground times can be analyzed.

7
and the flight’s scheduled arrival time is

ta2 = td2 + m2 + b2 = ta1 + tg + m2 + b2 = m1 + b1 + tg + m2 + b2 . (6)

As before, the actual arrival time of flight 1 is bta1 = m1 + 1 . The actual arrival time of
flight 2 equals bta2 = btd2 + m2 + 2 , where btd2 is the actual departure time of flight 2. To find
btd2 , note that if flight 1 is late in arriving, then the ground time will be reduced below the

scheduled time tg in an attempt to prevent delay propagation via late departure (and possible
late arrival) of flight 2. However, ground time cannot be reduced below the minimum feasible
time, equal to tg . Therefore, the actual departure time of flight 2 is given by

btd2 = max{m1 + 1 + tg , td2 } = max{m1 + 1 + tg , m1 + b1 + tg }. (7)

Flight 2 departs on time if 1 + tg < b1 + tg or

1 < b1 + tg − tg = b1 + bg . (8)

Note that satisfaction of (8) is ensured if flight 1 is early, or if 1 < b1 . But the inequality can
also be satisfied when flight 1 is late provided that it is not too late. Using (8), the probability
of an on-time departure for flight 2 is F1 (b1 + bg ), with late departure (which reverses the
inequality in (8)) having probability 1 − F1 (b1 + bg ).
The formula for btd2 allows derivation of conditions for late arrival of flight 2. Using (6),
flight 2 is late in arriving when

bta2 = btd2 + m2 + 2 > ta2 = m1 + b1 + tg + m2 + b2 . (9)

If flight 2 departs on time, so that btd2 = m1 + b1 + tg from (7), then (substituting in (9)), it
arrives late when

2 > b2 , (10)

8
an outcome that has probability 1 − F2 (b2 ) (early arrival occurs when (10) is reversed and has
probability F2 (b2 )). If flight 2 departs late, so that btd2 = m1 + 1 + tg , then (substituting in
(9)), it arrives late when

1 + 2 > b1 + b2 + bg , or

2 > b1 + b2 + bg − 1 . (11)

With late departure, flight 2 arrives early when (11) is reversed.

Table 1 summarizes the preceding information, while showing flight 2’s arrival delay. A
key observation from the table is that when flight 2’s departure is delayed, its arrival delay
depends on the sum b1 +b2 +bg . Therefore, the ground buffer affects delay by altering this sum,
conditional on a late departure for flight 2. However, bg also affects whether a late departure
occurs via the direction of the inequality in the second column of the table. As will be seen, the
overall impact of bg on the airline’s objective function operates through both these channels.
Using (11) and the reverse of inequality (8), the probability of late departure and late
arrival for flight 2 is given by
Z 1 Z 2
Pr(bta2 > ta2 ∩ btd2 > td2 ) = f1 (1 )f2 (2 )d2 d1 . (12)
1 =b1 +bg 2 =b1 +b2 +bg −1

Similarly, using the reverse of inequality (11) and the reverse of (8), the probability of late
departure and early arrival for flight 2 is
Z 1 Z b1 +b2 +bg −1
Pr(bta2 < ta2 ∩ btd2 > td2 ) = f1 (1 )f2 (2 )d2 d1 . (13)
1 =b1 +bg 2 =0

Combining all this information, the probability of a late arrival for flight 2 is given by

Pr(bta2 > ta2 ) =

Pr(bta2 > ta2 | btd2 = td2 ) Pr(btd2 = td2 ) + Pr(bta2 > ta2 ∩ btd2 > td2) =
Z 1 Z 2
[1 − F2 (b2 )]F1(b1 + bg ) + f1 (1 )f2 (2 )d2 d1 . (14)
1 =b1 +bg 2 =b1 +b2 +bg −1

9
Note that a conditional probability can be used in the first half of (14) because whether flight
2 is late conditional on an on-time departure (which depends only on 2 ) is independent of
whether the departure is on-time (which depends only on 1 ). The absence of this kind of
independence when flight 2 is late requires the different kind of expression in the last half of
(14).
Similarly, the probability of early arrival for flight 2 is given by

Pr(bta2 < ta2 ) =

Pr(bta2 < ta2 | btd2 = td2) Pr(btd2 = td2 ) + Pr(bta2 < ta2 ∩ btd2 > td2 ) =
Z 1 Z b1 +b2 +bg −1
F2 (b2 )[1 − F1 (b1 + bg )] + f1 (1 )f2 (2 )d2 d1 . (15)
1 =b1 +bg 2 =0

3.2. The airline’s objective function

The disutility for passengers of flight 1 is again given by in Ω1 in (2). The expected
disutility from late arrival of flight 2 is given by

Z 2
Ω2,late = xF1(b1 + bg ) (2 − b2 )f2 (2 )d2 +
b2
Z 1 Z 2
x [1 + 2 − (b1 + b2 + bg )]f1 (1 )f2 (2 )d2 d1 . (16)
1 =b1 +bg 2 =b1 +b2 +bg −1

The first half of (16) captures disutility from a late arrival when flight 2 departs on time (note
that 1 − F2 in (14) is replaced by the integral). The rest of (16) captures late disutility when
the departure is late (note that the bracketed term is added inside the integrand in (14)).
Similarly, the expected disutility from early arrival of flight 2 is

Z b2
Ω2,early = yF1 (b1 + bg ) (b2 − 2 )f2 (2 )d2
2
Z 1 Z b1 +b2 +bg −1
+ y [b1 + b2 + bg − (1 + 2 )]f1 (1 )f2 (2 )d2 d1 . (17)
1 =b1 +bg 2 =0

10
The passenger-disutility portion of the airline’s objective function, denoted by Ω, is given by
the sum of (2), (16), and (17):

Ω = Ω1 + Ω2,late + Ω2,early . (18)

In contrast to the single-flight case, our airline cost expression assumes the absence of
excess capacity in the two-flight case, with the airline adjusting its aircraft lease to exactly
cover the scheduled usage of the plane (see below for an alternate assumption). The airline’s
costs are then given by cf b1 + cf b2 + cg bg plus a constant involving the m0i s, with the carrier
paying for only the ground time allotted to the ground buffer and not for any time beyond
the termination of flight 2. The airline minimizes the sum of Ω and this cost expression, and
the derivatives of Ω with respect to bg , b1 and b2 are computed in the appendix. Adding the
relevant buffer cost to these derivatives and setting the resulting expressions equal to zero
yields the following first-order conditions:

∂Ω
bg : + cg =
∂bg
Z 1  
x
(x + y) F2 (b1 + b2 + bg − 1 ) − f1 (1 )d1 + cg = 0 (19)
1 =b1 +bg x+y
 
∂Ω ∂Ω x
b1 : + cf = + (x + y) F1 (b1 ) − + cf = 0 (20)
∂b1 ∂bg x+y
 
∂Ω ∂Ω x
b2 : + cf = + (x + y)F1(b1 + bg ) F2 (b2 ) − + cf = 0. (21)
∂b2 ∂bg x+y

Note in (20) and (21) that ∂Ω/∂b1 and ∂Ω/∂b2 are equal to ∂Ω/∂bg plus the second term in
the relevant equation.

3.3. Characterizing the optimum

The immediate implication of the first-order conditions is that the flight buffer b1 has the
same value as in the single-flight model. This conclusion can seen by using (19) to substitute
−cg in place of ∂Ω/∂bg in (20), which yields a condition that matches (3) from the single-flight

11
model. With flight 1’s buffer set as if the flight were operating in isolation, the buffer therefore
plays no role in mitigating delay propagation. Thus,

Proposition 1. Responsibility for mitigation of delay propagation falls only on the

ground buffer and the flight buffer for flight 2.

The lack of a delay-propagation role for flight 1’s buffer makes sense. Even though b1 and
the ground buffer are, in effect, perfect substitutes in addressing delay propagation, use of
the ground buffer instead of the flight buffer does not distort the balance between late and
early disutilities for flight 1, making it the preferred tool. However, suppose the ground buffer
were somehow constrained below its optimal value, due to a shortage of airport gate capacity,
for example, which could be caused by hoarding of gates by a dominant airline (see Ciliberto
and Williams (2010) for empirical evidence). In this case, b1 would help to address delay
propagation along with the other buffers. Such a constraint would make ∂Ω/∂bg in (20) less
than −cg , causing b1 to rise above its single-flight value, thus addressing delay propagation.
To see how the delay-propagation responsibility is apportioned between flight 2’s buffer
and the ground buffer, it is instructive to first consider the unrealistic case where cg = 0, with
ground time being costless. When cg = 0, F1 (b1 + bg ) = 1 must hold at the optimum, so
that the probability of late departure for flight 2 (which requires 1 > b1 + bg ) equals zero. In
effect, the lower limit of integration in (19) must be above the upper limit in order to make the
integral zero. To establish this point, suppose to the contrary that F1 (b1 + bg ) < 1 is satisfied
along with the first-order conditions. Since F2(b1 + b2 + bg − 1 ) ≤ F2 (b2 ) holds over the range
of integration in (19), which is nonempty given the maintained assumption, it follows that

 Z 1
∂Ω x
< (x + y) F2 (b2 ) − f1 (1 )d1
∂bg x+y 1 =b1 +bg
 
x
= (x + y) F2 (b2 ) − [1 − F1 (b1 + bg )] < 0, (22)
x+y

where the last inequality follows because F2(b2 ) − [x/(x + y)] < 0 by (21) (using ∂Ω/∂bg = 0
from (19)).

12
 The inequalities in (22) contradict the assumption that (19) equals zero, ruling out the
premise that F1 (b1 + bg ) < 1. Therefore, the ground buffer bg is set at a value large enough to
eliminate the chance of late departure for flight 2, so that F1 (b1 + bg ) = 1.7
Then, setting F1(b1 + bg ) in (21) equal to 1, (21) matches the optimality condition (3) for
the single-flight model when cg = 0, implying that the optimal b2 equals the single-flight value.
Thus, when cg = 0, the flight buffer for flight 2 is set as if the flight were operating in isolation,
with delay propagation not an issue. Since the same conclusion has already been established
for flight 1, we can state

Proposition 2. When ground time is costless, the ground buffer does all the work in
mitigating delay propagation, fully eliminating it, with no contribution from flight 2’s
buffer.

Even though this is a natural conclusion, some work is required to derive it from the model,
as seen in the previous discussion.
Now consider the realistic case where the cost of ground time is positive, with cg > 0.
Setting (19) equal to zero now implies that the integral must be negative, which means that
the lower limit of integration cannot exceed the upper limit as before, leading to a zero value
for the integral. In other words, F (b1 + bg ) must now be less than rather than equal to 1,
indicating that there is a chance of late departure for flight 2.
In this case, the mitigation of delay propagation is apportioned between the ground buffer
and flight 2’s buffer, with the exact apportionment depending on the relationship between cf
and cg . Suppose first that cf < cg , so that the cost of the flight buffer is less than that of the
ground buffer. Letting ** denote optimal values in the two flight model, it follows from (20)
that F2 (b∗∗ ∗∗ ∗∗ ∗∗
2 )−x/(x+y) > 0 must hold. Since F1 (b1 +bg ) < 1, (21) and F2 (b2 )−x/(x+y) > 0

imply
 
x
0 = (x + y)F1 (b∗∗
1 + b∗∗
g ) F2 (b∗∗
2 ) − + cf − cg
x+y
7 It should be noted that F (b + b ) = 1 does not yield a unique solution for b given that any b value
1 1 g g g
satisfying b1 + bg ≥ 1 makes the equality true. However, replacing b1 by b∗1 , the airline might be assumed
to choose the smallest bg satisfying the equality, so that b∗1 + bg = 1 , yielding a unique solution given by
b∗g = 1 − b∗1 .

13
  
x
< (x + y) F2(b∗∗
2 ) − + cf − cg (23)
x+y

(recall b∗∗ ∗ ∗∗ ∗
1 = b1 ). With the last line of (23) thus positive, it follows that b2 is larger than b2 ,

the single-flight value of b2 , which makes the second line equal to zero. Since flight 2’s buffer
is thus larger than the value it would assume if the flight were operating in isolation, it follows
that flight 2’s buffer assists the ground buffer in addressing delay propagation. This conclusion
is due to the relative cheapness of the flight buffer, which encourages its use in addressing
propagation.
In the reverse case where cf > cg , reversal of the above argument yields b∗∗ ∗
2 < b2 , so that

flight 2’s buffer is less than its single-flight value. Now, relative cheapness of the ground buffer
means that it takes extra responsibility in addressing delay propagation, causing flight 2’s
buffer to be reduced below its single-flight value. This adjustment means that the flight buffer
now actually contributes to delay propagation, but this effect is offset by the greater role of the
ground buffer. Finally, when cf = cg , flight 2’s buffer equals its single-flight value (b∗∗ ∗
2 = b2 ),

so that it neither helps nor offsets the ground buffer in mitigating delay propagation. With
the buffers equally costly, the nondistorting ground buffer is thus set to do all the work in
addressing delay propagation, although the chance of propagation is not reduced to zero given
the costliness of the buffer. Summarizing yields

Proposition 3. When cg > 0, mitigation of delay propagation is apportioned between

the ground buffer and flight 2’s buffer. When cf = cg , the ground buffer alone addresses
delay propagation (while not fully eliminating it), with flight 2’s buffer set at its single-
flight value. When cf < (>) cg flight 2’s buffer contributes to (partly offsets) the ground
buffer’s mitigation of delay propagation, taking a value above (below) its single-flight
value.

The results yield a further implication in the case where the distributions of the random
flight-duration terms are equal, allowing the 1 and 2 subscripts to be removed from the F
functions. With common F ’s, the single-flight values of b1 and b2 are the same. Then, the fact
that b2 is greater than (less than) the common single-flight value as cf < (>) cg means that
flight 2’s buffer is greater than (less than) flight 1’s buffer, which always equals the common
single-flight value, when cf < (>) cg . In other words, b∗∗ ∗∗
2 > (<) b1 holds as cf < (>) cg . The

14
reason is that flight 1’s buffer plays no role in addressing delay propagation, while flight 2’s
buffer contributes to (borrows from) the ground buffer’s delay-mitigation effect as cf < (>) cg .
The second-order conditions for the optimization problem have not been mentioned so far.
In the appendix, it is shown that, if cf ≥ cg , then Ω is strictly convex in b1 and b2 at the
optimum, so that conditional on bg , flight buffers satisfying the first-order conditions based on
(20) and (21) yield a local minimum of Ω. However, it is not possible to establish convexity of
Ω in all three buffers, a condition that must be assumed to hold.
With b1 always equal to its single-flight value, comparative-static results for this flight
buffer are given by (5). Comparative statics for b2 and bg are generally unavailable, however,
because they require total differentiation of the equation system (19)–(21), which leads to
ambiguous results.
If the current model were extended to a case where the aircraft provides three or more flights
per day, greater complexity would rule out derivation of analytical results, with reliance on
numerical simulation required instead. Such analysis, however, would undoubtedly reproduce
one of the main current results by showing that first flight’s buffer equals its single-flight value.
It is hard to predict, however, what the rest of the numerical analysis would show.

3.4. The effects of adding or removing excess capacity

As seen above, the single-flight model assumes that excess capacity is present, while it is
absent in the two-flight model. These assumptions affect the comparison between flight buffers
in the two models, making it useful to gauge the effect of alternate assumptions.
First, suppose that excess capacity is present in the two flight model, in which case the
previous airline cost expression is replaced by cf b1 + cf b2 + cg (T − b1 − b2 ), where T is the total
time covered by the aircraft lease. Ground time now includes the ground buffer and additional
ground time following the termination of flight 2. It is easy to see that, following this change,
cg disappears from the bg first-order condition (19), whereas cf in (20) and (21) is replaced by
cf − cg . Arguments parallel to those leading to Proposition 2 can then be applied, yielding the
same conclusions as in the proposition, namely, sole usage of the ground buffer in addressing
delay propagation. Intuitively, when it comes to choice of the ground buffer, excess capacity
is just like a zero cg , accounting for the identical conclusions. Note, however, that a corner

15
solution emerges if T is not large enough to accommodate a sufficiently long ground buffer.8
Alternatively, suppose that excess capacity is absent in both the single- and two-flight
models. Then, the single-flight cost expression becomes cf b1 , and −cg vanishes from (3). This
change makes the expression in (3) larger, which requires a decline in the value of b1 in order
to maintain the zero equality. With the single-flight value falling and the flight buffers in the
two-flight model unchanged, b∗∗ ∗ ∗∗ ∗
1 > b1 now holds, while b2 could be larger or smaller than b2

when cf > cg . Being larger than the single flight-value, flight 1’s buffer therefore now assists
the ground buffer in mitigating delay propagation.
While these conclusion are not unreasonable, we believe that the results in Propositions
1, 2, and 3, which are based on excess capacity in the single-flight model, are more natural
and intuitive. Moreover, when only one flight is operated, the existence of excess capacity is a
plausible assumption.

4. Numerical examples
Figures 2–5 present numerical examples. It should be noted that, given the stylized nature
of the model, realism in the choice of parameter values is not possible, and the qualitative
(rather than quantitative) effects of parameters changes are of interest.9
Since some of the comparative-static effects of x and y and of cf and cg have been derived
analytically, the analysis focuses on the effects of greater variability in the random terms 1 and
2 . To generate the results, 1 and 2 are assumed to follow independent normal distributions
with a positive mean µ. Their standard deviations start out equal, satisfying σ1 = σ2 = 0.0.
Then, each of the σ’s increases up to 1.5 holding the other σ fixed, allowing the effect of greater
flight-time variability for flights 1 and 2 to be appraised separately. Next, the σ’s are set at a
common value and increased simultaneously from 0.0 up to 1.5, allowing the effects of greater
overall flight-time variability to be appraised.
Among the other parameters, the minimum turnaround time tg is set at 0.5 hours, and

8 This discussion shows that, as in the c = 0 case, unrealistic conclusions make excess capacity an undesirable
g
assumption for the two flight-model.
9 He (2019) uses such numerical exercises involving two flight legs and two aircraft to illustrate an integrated
approach for overbooking and capacity planning.

16
the lateness and earliness disutilities are set at x = 1.0 and y = 0.1, with y realistically much
smaller than x. The buffer costs are initially set at cf = 0.05 and cg = 0.01.
Recalling that the ’s are assumed to be positive random variables, this property can be
assured by choosing the normal mean µ to be sufficiently large. With σ ranging from zero
up 1.5, the probability of a negative  is smaller than 0.025 when the mean is at least twice
the maximum σ, or at least 3.0, and its value is assumed to lie far enough above 3 to yield
a negligible probability. Furthermore, inspection of the first-order conditions (19)–(21) shows
that raising the value of the mean by ∆µ (which shifts the distribution function F to the right
by this amount) increases both flight buffer solutions by ∆µ while leaving the ground buffer
solution unchanged. Thus, the flight buffers move exactly in step with µ. This fact is used to
facilitate presentation of the following figures, allowing flight and ground buffers to be shown
on the same scale. In particular, the flight-buffer values shown are based on a µ value of zero,
and they must be incremented by the amount of the actual µ (which is above 3) to get the
actual solutions. Since σ’s effect on the buffers is our main focus, this need for rescaling is
inconsequential.10
Figure 2 shows the effect of increasing σ1 from 0.0 to 1.5 with σ2 fixed at 0.5. As can be
seen, the flight buffer b1 rises as σ1 increases, a natural finding, while the ground buffer bg
also rises. The buffer b2 for flight 2 appears to be constant in the figure, but it increases very
slightly with σ1 . The conclusion, therefore, is that b1 and bg alone do almost all the work in
absorbing the greater chance of an arrival delay and subsequent delay propagation that follows
from an increase in flight-time variability for flight 1.
Figure 3 shows the effect of increasing σ2 from 0.0 to 1.5 with σ1 fixed at 0.5. Now b2
rises, while b1 is constant (note that the b1 solution from (25) is independent of σ2 ). However,
the ground buffer bg is decreasing in σ2 , apparently because greater flight-time variability for
flight 2 makes the ground buffer less effective at preventing a late arrival.11
10 For computational reasons, the infinite upper and lower limits of the normal distributions are replaced by
10 and −10 respectively in the calculations. With σ’s taking the values mentioned above, the probability that
1 or 2 lies outside this range is virtually zero, making the restriction inconsequential.
11 Alternatively, recall that, conditional on delay propagation, the second flight’s arrival delay depends on the
sum of buffer times b1 + b2 + bg (see Table 1). If the second flight’s buffer increases because of an increase in
σ1 , the result is an increase in the sum of buffer times. The ground buffer is then reduced to moderate the

17
 Figure 4 shows the effect of simultaneously increasing σ1 and σ2 from 0.0 to 1.5. Both
flight buffers naturally increase with the common σ value, and although the figure makes the
buffers look equal in size, b1 is slightly larger than b2 , as predicted. In addition, bg is increasing
in the common σ value. However, Figure 5 shows that the behavior of the ground buffer is
reversed when cf = 0.05 and cg = 0.03, falling as the common σ value increases (matching
the outcome in Figure 2). Figure 5 shows an additional point that does not arise in the other
cases. In particular, bg becomes negative as σ increases, showing that the ground time is set
below the minimum feasible turnaround time tg . Nothing in the model prevents this outcome,
which need not lead to late departure for flight 2 if flight 1’s buffer is sufficiently large. In the
data discussed below, however, the outcome is exceedingly rare, accounting for only 0.2% of
the observations.
The implication is that the effect of flight-time variability on the ground buffer could be
positive or negative depending on the magnitudes of the other parameters. If the ground buffer
is sufficiently cheap compared to the flight buffers (cg = 0.01 vs. cf = 0.05), it is used along
with the flight buffers to address the greater threat of delay propagation resulting from higher
flight-time variability (Figure 4). But when the ground-buffer cost is larger as a proportion of
cf (cg = 0.03 vs. cf = 0.08), then the flight buffers partly supplant the ground buffer as the
threat of delay propagation rises, with bg falling (Figure 5).

5. Adding passenger connections

So far the analysis has suppressed the possibility that some passengers connect from flight 1
to flight 2. These passengers would travel from the origin city of flight 1 to flight 2’s destination
city, making a connecting trip in the absence of nonstop service between the two cities. This
type of connecting travel, however, has little effect on the model. Assuming that a share α
of passengers on both flights are traveling nonstop while 1 − α are connecting, the airline’s
objective function would be altered in a straightforward way. In (18), Ω would be multiplied
by α and then added to the term (1 − α)(Ω2,late + Ω2,early ), which represents the late and
early disutilities for connecting passengers, who care only about their arrival time at flight

increase in this sum.

18
2’s destination. Adding the two expressions, the disutility portion of the objective function
reduces to (18) with α multiplying Ω1 .
A more interesting and complex connecting scenario arises if two additional flights, again
using a single aircraft, are added to the model. The turnaround city for these flights, which
are denoted 1B and 2B, is the same as for the original flights, now denoted 1A and 2A. This
common turnaround city can thus be viewed as a hub, which is a destination in its own right
but also supports passenger connections. Connecting passengers now include those traveling
from flight 1A’s origin (OA ) to flight 2B’s destination (DB ) as well as those traveling from OB
(flight 1B’s origin) to DA (flight 1A’s destination).12
If flight 1A arrives after the departure of flight 2B, passengers traveling from OA to DB miss
their connection, suffering disutility V , with same conclusion applying to passengers connecting
from flight 1B to flight 2A. Note that since connecting passengers using flights 1A and 2A (or
1B and 2B) do not change planes, a missed connection is not possible for them.13
The portion of the airline’s objective function applying to nonstop trips and same-plane
connecting trips is given by adding the Ω’s for the A and B flights, with the previous α mod-
ification incorporated.14 The part of the objective function that applies to the remaining
connecting passengers makes use of the probability of a missed connection. For 1A-2B connec-
tions, this probability is PAB ≡ P r(m1 + 1A > btd2B ), using (7) and adding an A subscript,
while for 1B-2A connections, the probability is given by the analogous expression PBA . In the
previous expression, m1 + 1A is the arrival time of flight 1, and a missed connection occurs
when it is greater than flight 2’s departure time.
Using PAB and PBA , the remaining (disutility) part of the objective function is given by
1 − α times

PAB V + (1 − PAB )(Ω2B,late + Ω2B,early ) + PBA V + (1 − PBA)(Ω2A,late + Ω2A,early ). (24)

12 Symmetry holds, with corresponding flight distances equal and with flights 1A and 1B departing at a
common time.
13 Note that, while crew members traveling as passengers to their next flight assignment can also miss con-
nections, with serious flight-delay consequences, analysis of this phenomenon would be complex.
14 The relevant expression is αΩ + Ω
1A 2A,late + Ω2A,early + αΩ1B + Ω2B,late + Ω2B,early . Note that the buffers
inside the Ω expressions also acquire A and B subscripts, although their equilibrium values will be the same
given symmetry.

19
The first and third terms in (24) give disutilities from missed connections, while disutilities for
connecting passengers who make, rather miss, their connection are given by the second and
fourth terms. Note that these early-late disutilities are the same as for nonstop passengers on
either flight 2A or 2B, being given by the terms multiplying 1 − PAB and 1 − PBA .
These second and fourth multiplicative terms in (24) make the objective function consid-
erably more complex than for a model with only two flights. Given the challenging nature of
resulting analysis, persuing it is beyond the scope of the paper. Intuitively, however, avoid-
ance of missed flight connections provides an additional reason beyond mitigation of delay
propagation to increase the flight buffers for flights 1A and 1B as well as the corresponding
ground buffers. Despite the absence of concrete conclusions beyond this simple intuition, it
is interesting nevetheless to see the logic under which flight connections can be added to the
model.15

6. Empirical Setup
6.1. Predictions
The empirical work aims to test some of the predictions of the theoretical model, using
US data that tracks the flights of individual commercial aircraft for 2018. The data allow
computation of the flight buffer for a particular flight, which is set equal to scheduled flight
time minus the minimum observed flight time on the route, matching the model. The data
also allow computation of the ground buffer, which equals scheduled ground time minus the
minimum observed ground time at the airport (details on both calculations are presented
below).
One of the model’s predictions, which comes from the simulation analysis of the two-flight
case, is that the flight buffer increases when a flight’s own time variability rises, but that the
buffer is unaffected by higher time variability for the other flight (Figures 2 and 3). Also,
the ground buffer is increasing (decreasing) in the time variability of the previous (subsequent)
15 A different model exploring missed connections would proceed as follows, returning to the two-flight frame-
work but assuming the flights use different aircraft (so that delay propagation is absent). If flight 1 arrives
after the departure of flight 2, then connecting passengers miss their connection, again generating disutility V .
Now, there is no ground buffer per se, but the airline chooses flight 2’s scheduled departure with an eye toward
avoiding missed connections, assuming that a later departure imposes the same costs as a longer ground buffer.

20
flight. To test these predictions, flight-time variability is measured by the standard deviation of
actual flight times at the flight, route and month level. Later regressions replace this variability
measure with proxies for variability.
A second prediction is that buffer costs affect the levels of all three buffers, although the
effect is unambiguous only in the case of b1 , which is decreasing in cf and increasing in cg . We
use airport characteristics (mainly hub status) as a proxy for cg and aircraft size as a proxy
for cf .
A third prediction is that the buffers depend on the disutilities x and y for lateness and
earliness, with unambiguous predictions only for available for b1 , which is increasing in x and
decreasing in y when cf and cg are close in magnitude. We use the managerial employment
share of the origin city as a proxy for x on the belief that such employees are highly averse to
lateness.
A fourth prediction is that the position of a flight in the day’s flight sequence affects the
flight buffer. Recall that flight 2 has a longer buffer than flight 1 when cf < cg , with the
relationship reversed when cf > cg . To test for such a sequencing effect, we include time-of-
day departure variables (morning, evening, etc.) along with variables that measure a flight’s
exact position in the sequence of an aircraft’s daily flights.
A fifth prediction is that the share of connecting passengers affects the levels of the buffers.
No explicit predictions are available, but we expect the ground buffer to increase as the con-
necting share on the subsequent flight rises. The connecting share is computed at the route-
airline-quarter level for the prior year using U.S. Department of Transportation (DOT) data.

6.2. Data collection and buffer measurement

The sample is obtained from the DOT and covers the US domestic airline market for the
year 2018. We mainly rely on the ‘Marketing Carrier On-Time Performance’ dataset, which
for each aircraft, uniquely identified by its tail number, contains information on the carrier
operating the flight, the origin and destination, the departure date, the scheduled departure
and arrival times, the actual departure and arrival times, and the taxi-in and taxi-out times.16
16 The novelty of this dataset relative to the ‘Reporting Carrier On-Time Performance’ dataset, previously used
in Forbes (2008) and other studies, is that it distinguishes whether the flight is operated by the reporting carrier

21
We exclude international flights, flights that are canceled or diverted, and flights from/to US
Commonwealth areas and Territories.
The aim of our empirical analysis is to investigate the determinants of the two choice
variables described in our theoretical model: the flight buffer and the ground buffer. The flight
buffer is obtained as follows. First, for any flight i, where i defines the sequence of daily flights
operated by a given aircraft during the day, we measure the actual flight time, which is the sum
of the taxi-out, airborne and taxi-in times (equivalent to the so-called ‘block-time’). Using the
b i = btai − btdi, the difference between flight
model notation, this actual flight time is given by m
i’s actual arrival and departure times. We then take the smallest value of m
b i across all aircraft
of flight i’s type flying the same route to obtain m
b min , the minimum flight time by route and
aircraft type (the route and type subscripts are suppressed for simplicity). The flight buffer is
given by the difference between flight i’s scheduled flight time mi and minimum flight time of
b min .17
the same aircraft type flying the same route: bi = mi − m
To calculate the ground buffer, we first compute the actual ground time that separates
flight k and flight k + 1 in the sequence of flights operated by the observed aircraft during
the day: btkg = btd,k+1 − btak . Then we calculate the minimum turnaround time as the shortest
actual ground time observed across all aircraft of a given type using the turnaround airport.
The ground buffer for an aircraft turnaround is obtained by subtracting this minimum feasible
ground time from the scheduled ground time.18 In order to have confidence in the accuracy
of the observed minimum flight and ground times used in the buffer computations, we require
that the number of observations used to generate them is at least equal to 30.

or by its regional codeshare partners. Both datasets are downloadable at www.transtats.bts.gov/Tables.asp

?DB ID=120&DB Name=Airline%20On-Time%%20Performance%20Data&DB Short Name=On-Time. However the
Marketing Carrier On-Time Performance dataset only starts with January 2018, while the Reporting Carrier
On-Time Performance dates back to 1987.
17 Changing this calculation so that the minimum flight is computed by route-carrier-quarter-airline instead
of by route-carrier has little effect on the results presented below. Our use of the minimum flight time follows
Mayer and Sinai’s (2003) use of the minimum to compute flight delays. Other papers instead calculate delays
using a low-percentile flight duration such as the 5th, 10th or 20th percentile (see Britto, Dresner and Voltes
(2012) and Zou and Hansen (2012).
18 Since the Marketing Carrier On-Time Performance dataset does not include international flights, except
those from/to US Commonwealth areas and Territories, we restrict the ground buffer to be not greater than 200
minutes, to exclude possible incomplete records that may occur when an aircraft flies from/to a non-domestic
destination (e.g. Canada or Mexico) between two domestic flights.

22
 We initially restrict the analysis to those aircraft that only operate two flights a day, as in
the model, using what we refer to as the “two-flight sample”. This restriction predominantly
limits the focus to coast-to-coast flights or flights to/from Alaska. However, we later present
results using an expanded sample that includes aircraft flying as many as 8 flights in a day,
referred to as the “unrestricted sample.”

6.3. Empirical approach

Our empirical analysis is based on two sets of regressions: one for flight buffers and another
for ground buffers. The standard errors are clustered by route and month in order to allow the
residuals of different aircraft (possibly of different carriers) flying on the same route, during
the same month, to be correlated.
The general equation to be estimated is:

Bufferjcodt = Xjcodt β + ηc + φo + ρd + λt + ujcodt (25)

where Buffer is either the flight buffer or the ground buffer. The subscript j identifies the
aircraft tail number, c the carrier, o the origin airport, d the destination airport (or the
turnaround airport in the ground-buffer regression), and t the month. The ηc , φo, ρd , and
λt terms are carrier, origin, destination, and month fixed effects19 and ujcodt is the regression
error, assumed i.i.d. with zero mean. The other independent variables are denoted by X, and
they control for different aspects of the buffer decision.20
Table 2 shows definitions and summary statistics for the variables in the two-flight and
unrestricted samples. In addition to the buffer, time-variability, and connecting-share variables
19 The airport fixed effects are not only relevant in the flight-buffer regression, but also in the ground-buffer
regression. Consider for example two flights that share the same turnaround airport and aircraft type, but
that originate from different airports. These flights have the same minimum actual ground time, but they may
have different load factors depending on the strength of demand and other forces. The potential difference in
the load factor may affect the turnaround time and hence the ground buffer set by the airline. The inclusion of
originating airport fixed effects (the origin airport of the previous flight) aims to control for this unobservable
characteristic.
20 Note that the regression in (25) bears some resemblance to scheduled-block-time regressions like those
estimated by Kang and Hansen (2017, 2018) and others. With the flight buffer equal to scheduled-block-time
minus minimum flight time, (25) would become a scheduled-block-time regression if minimum flight time were
moved to the RHS and given a coefficient. Kang and Hansen’s regressions, however, differ from this modified
regression by not including minimum flight time as a covariate.

23
discussed above, the table shows the month and airline dummies from (25). Other variables
included in the set of X variables from (25) are hub origin and hub destination dummies
(replaced by a hub turnaround variable for the ground-buffer regressions), congestion measures
for the origin and destination (or turnaround) airports (equal to traffic divided by runway
capacity),21 a slot-control dummy for the turnaround airport, dummies for regional and low-
cost carriers, a distance measure, time-of-day and weekend dummies, two variables indicating
a flight’s position in the aircraft’s daily sequence, a large-aircraft dummy, and the managerial-
share variable (for the origin or turnaround airport).
In the unrestricted sample, the mean flight buffer is about 32 minutes, indicating that
carriers set scheduled flight times to be 32 minutes longer on average than the minimum flight
time for that route and aircraft type. The mean ground buffer shows that airlines set ground
times to be 38 minutes above minimum turnaround times. The mean connecting share is
45%,22 36% of flights have a hub origin or destination, 31% of flights are on regional carriers,
29% are on low-cost carriers, routes have one competitor on average, average distance is a
bit over 800 miles, 27% of flights are on weekends, 3% of flights use heavy aircraft, and the
managerial share of the origin work force averages about 5%. Differences in means between
the samples are mostly as expected (for example, mean distance more than doubles to nearly
1700 miles moving to the two-flight sample).

7. Regression results
7.1. Flight-buffer regressions with flight-time variability
The results of the flight-buffer regressions including flight-time variability are shown in
Table 3. Many of the variables listed in Table 2 are omitted from these regressions (to be
included later) on the grounds that they may be proxies for flight-time variability, thus being
21 The data on number of runways for the sample airports are from the FAA’s National Flight Data Center,
available at https://www.faa.gov/air traffic/flight info/aeronav/aero data/Airport Data/.
22 The connecting shares are computed using ticket data from the DOT’s Airline Origin and Destination
Survey (database DB1B), which comes from quarterly 10% sample of all tickets. For each route-airline-quarter
combination in 2017, the share of connecting passengers on the route can be deduced from the DB1B ticket
information. This share is then matched to the route-airline-quarter observations in our data, thus indicating
the average share of connecting passengers present for each of our observations. While some routes carry zero
connecting passengers in actuality, other routes may show zero connections because the sampling process misses
small numbers of passengers.

24
excludable given that variability is explicitly measured. In addition to other variables in Table
2, the excluded variables include the month dummies and the origin and destination fixed
effects, all of which are likely determinants of flight-time variability. As mentioned above, the
variability measure used in the regressions is the standard deviation of the actual flight times,
computed by route, month and flight number.
In order to match the setup of our theoretical model, the regressions in Table 3 are based
on a subsample of aircraft that operate only two flights a day (mainly flying coast-to-coast
or to Alaska). We refer to this sample as ‘two-flight sample’, and it is expanded in later
regressions to include aircraft operating more flights per day . In the table, column (1) shows
the regression with the flight buffer of flight 1 set as dependent variable, column (2) has flight
2’s flight buffer as dependent variable, and columns (3) and (4) pool the two flights, with (4)
adding the dummy variable Flight 2 to identify the second flight. Each regression includes
flight variability for both flights, following the simulations.
The time-variability coefficients are positive and significant in all four regressions. Columns
(1) and (2) show that an increase in either flight’s time variability leads an airline to raise the
flight buffer for both that flight and the other flight. This finding does not perfectly match
the simulations, which showed that increasing σ for just a single flight has a small or zero
effect on the other flight’s buffer. Columns (3) and (4) show that when the flights are pooled,
time variability for both flights affects the buffers for either flight, yielding almost the same
conclusion as columns (1) and (2). Despite this imprefect correspondence, the findings confirm
the overall spirit of the numerical examples, which show that greater time variability tends to
raise flight buffers.
The share of connecting passengers (measured for flight 2) has a significantly negative effect
on the flight buffers in all four regressions, an unexpected result. A more intuitive finding
emerges, however, in the ground-buffer regressions presented next. The dummies indicating
a hub origin or destination are mainly meant to capture the cost of ground time, with gate
rentals likely to be more costly at hub airports given their large sizes. The hub coefficients are
uniformly positive and significant, indicating that flights in or out of hubs have longer flight
buffers, an effect that may also reflect aversion to possible disruption of flight connections.

25
 The coefficients of the carrier dummies show long flight buffers for Delta and Virgin Amer-
ica (relative to American, the default carrier) and short ones for Allegiant, Frontier, Hawaiian,
Jet Blue, Southwest, Spirit. While the buffers of some of the latter carriers are shorter than
American’s by more than 10 minutes, United’s flight buffers are also seen to be shorter, but
by less than two minutes. Regional airlines in the dataset have lower flight buffers (by about
7 minutes) relative to other carriers, possibly indicating that these airlines are keen to max-
imize aircraft utilization. In addition, the uniformly positive and significant coefficients for
the competitors variable suggest that more competition may heighten an airline’s attention to
on-time performance, leading to an increase in its flight buffers. An extra competitor increases
flight buffers by over a minute. Routes with a high managerial share at the origin have longer
buffers, possibly reflecting business passengers’ aversion to late arrivals. Finally, flight buffers
are unexpectedly longer for large aircraft, which have higher values of cf . Column (4) of the
table shows a slightly shorter buffer for flight 2 relative to flight 1, whereas column (3) shows
a mixed pattern of time-of-day coefficients.23

7.2. Ground-buffer regressions with flight-time variability

Table 4 presents the ground-buffer results with flight-time variability, again using the two-
flight sample. Since the variability coefficients for the ground buffer are somewhat sensitive
to how variability is computed, we use two approaches. In columns (1) and (2), variability is
the 2017 standard deviation of flight times computed by route, month and flight number, as
before, whereas variability is computed only by route and flight number in columns (3) and
(4), not controlling for month.
Greater time variability for flight 2 reduces the ground buffer in all four regressions, match-
ing the results in Figure 3. The effect of flight 1’s time variability is insignificant in the first
two columns but significantly positive in columns (3) and (4), matching the results in Figure 2.
Thus, Table 4 shows that greater time variability for flight 2 (flight 1) reduces (raises or leaves
constant) the ground buffer, in remarkably close correspondence to the simulation results.
Turning to the other covariates, the coefficients of the share of connecting passengers are
23 Note that inclusion of the time-of-day variables is inappropriate in the regressions of columns 1 and 2
because, for example, flight 2 cannot generally be a morning flight.

26
uniformly positive and significant across the regressions, as expected. The point estimates
indicate that a 0.1 increase in the connecting share on flight 2 raises the ground buffer by
about 1 minute. The hub turnaround variable also has a positive and significant coefficient,
and since the connecting share is held constant, this variable should capture the cost of ground
time, with a negative coefficient expected. The positive estimate thus suggests that the hub
turnaround variable may also help to capture the presence of connecting passengers. Note that
a hub turnaround raises the ground buffer by about 13 minutes.
Among the airline dummies, notable results are the long ground buffers of Virgin America,
Hawaiian, United and Alaska, and the short buffers of Allegiant. The Southwest coefficient,
while significantly negative, shows a buffer only 2 minutes shorter than the reference carrier
(American). But when all the low-cost carriers are represented by a common dummy, the
variable’s significant coefficient shows ground buffers about 8 minutes shorter than non-LCCs.
Thus, recalling the short LCC flight buffers from Tables 3 and 6, LCCs keep their buffers
short both in the air and on the ground. Columns (1) and (3) show that regional carriers have
slightly shorter ground buffers than non-regional airlines.
The coefficients of the heavy-aircraft variable are insignificant, a finding that is reasonable
given that ground buffers are already type-specific. The managerial-share coefficient is now
significantly positive, indicating that ground buffers are longer in turnaround cities with a high
managerial share, in an attempt to secure an on-time departure for flight 2.
The time-of-day coefficients show that the ground buffer monotonically increases as the
turnaround between the two flights occurs later in the day. This effect is not captured in the
model, but it could indicate that, when the two-flight sequence starts later in the day, the
effects of time-varying congestion or other factors make a longer ground buffer optimal.

7.3. The determinants of flight time variability

The next step in the empirical analysis is to run regressions where flight-time variability is
replaced by a host of variables that help to determine variability. The regressions thus include
the previous variables along with proxies for variability but not variability itself, providing a
different and perhaps more revealing picture of the factors affecting flight and ground buffers.
In this exercise, we present results for both the two-flight sample and a larger “unrestricted”

27
sample described below.
Before showing these regressions, it is useful to explore the relationship between flight-
time variability and the set of proxy variables. These variables are the month dummies, which
control partly for weather conditions, the origin and destination congestion variables (which
presumably influence variability), the distance variable, the weekend dummy, and the origin
and destination dummies, whose coefficients are not reported.
The results are shown in Table 5 for the two-flight sample, using a separate regression for
each flight. With t statistics five to ten times those of other covariates, distance has a very
precisely measured positive impact, showing that longer flights naturally have greater time
variability. The congestion coefficients are also mostly positive and significant, as expected,
while weekend flights show low time variability. The month dummies somewhat surprisingly
indicate that the good-weather months of July and August have higher flight-time variability
than the Winter months. This outcome may be due to high summer travel volumes, which can
generate delays from various sources.24

7.4. Flight-buffer regressions with variability proxies

The results of the flight-buffer regressions with variability proxies are presented in Table
6. The regressions in columns (1)–(4) are based on the two-flight sample and have the same
structure as those in Table 3. The regressions in columns (5)–(7) remove the restriction to
aircraft operating just two flights per day. The corresponding sample, which we refer to as
the ‘unrestricted sample’, consists of aircraft that operate at most 8 flights in a day.25 Other
differences between the regression specifications are discussed below.
Where statistically significant, the monthly dummy coefficients have negative signs, in-
dicating that compared to January, the reference month, flight buffers tend to be shorter in
other months. These results point to a role for weather in influencing the flight-buffer deci-
24 It is worth pointing out that if flight-time variability is computed only by route and flight number, then
the month dummies become statistically insignificant. This result is mainly technical: because the dependent
variable is then invariant throughout the year, the month dummies do not explain much more than the constant
of the regression would explain, making their coefficients insignificant.
25 The maximum number of flights operated by a single aircraft observed in the ‘Marketing Carrier On-Time
Performance’ database is 16. However, since operation of 9 or more flights is seldom observed, representing less
than 0.40% of the initial database, these cases are not likely to meet our criterion of at least 30 observations
in the calculation of mb min and are therefore excluded from the unrestricted sample.

28
sion. The estimated coefficients suggest that the size of the effect varies with expectations of
bad whether: the flight buffer falls monotonically from March/April until September, which
is a period of the year generally characterized by good weather conditions and therefore by
fewer weather-related delays. The flight buffer then starts increasing monotonically, while still
remaining below the January value, until the end of the year. The statistically insignificant
coefficients on the February dummy and sometimes the March dummy mean that there is
no difference in flight-buffer choices across the months of January, February and (partially)
March. This result makes sense because January, February and March are winter months that
bear the same bad-weather expectations and therefore the same delay concerns.
In the two-flight sample, the effect of weather (in columns (3) and (4)) yields as much as
6 minutes of reduced buffer length during the Spring and summer months. In the unrestricted
sample, this effect is still present, but with a slightly smaller impact, at about 5 minutes
maximum. It should be noted that these month effects cannot be attributed to flight-time
variability given the results of Table 6. If weather effects on flight buffers operated only through
flight-time variability, then with a positive buffer-variability link, the month coefficients in
Table 6 would follow the pattern seen in Table 5. The contrary results in Table 6 thus appear
to show that the effect of weather may not operate entirely though flight-time variability.
Instead, bad weather may slow all flights (requiring longer buffers) without making flight
times more variable.
As in Table 3, the effect of a hub origin or destination is positive, although the origin
effect is insignificant in the unrestricted sample. Note that the coefficients of the connecting
share, rather than being significantly (and inexplicably) negative as before, are now uniformly
insignificant. The origin and destination congestion measures, which were not present in Table
3, have uniformly positive and significant coefficients, indicating that congestion at either
endpoint raises flight buffers. Recalling from Table 5 that time variability is positively related
to congestion, this positive congestion effect may thus operate through variability.26
The carrier coefficients are qualitatively similar to those in Table 3, although the absolute

26 Recall that our measure of flight time is block time, which includes the taxi time along with the airborne
time.

29
magnitudes tend to be smaller. The regional carrier dummy now has positive coefficients in
the unrestricted sample, reversing the previous negative sign (now possibly reflecting feeder
carriers’ concerns about missed connections at hubs). While the competition effect remains
positive, the large-aircraft dummy coefficients all flip sign to negative and significant, a welcome
shift that confirms expectations. The new distance variable has a uniformly positive effect on
flight buffers, and since distance and time variability are strongly related, this effect presumably
operates through variability. The effect is about one extra minute of buffer per hundred miles
flown. As for weekend flights, the weekend dummy coefficients in the two-flight sample are
negative, although the unrestricted sample shows a significantly positive effect. Note that the
managerial-share variable does not appear in Table 6 given its collinearity with the airport
dummies.
The time-of-day dummy coefficients in columns (3) and (5) show a clearer pattern than in
Table 3, with the buffer magnitudes higher late in the day than in the default early-morning
period. Matching this pattern, the flight-2 dummy coefficient is now significantly positive
rather than negative.
To explore the time-of-day effect in a different way, we investigate the effect on its buffer
of a flight’s position in the day’s flight sequence, using either flight-sequence dummies or the
aircraft-rotation variable (see Table 2).27 Column (6) of Table 6 shows results using flight-
sequence dummies in place of the time-of-day dummies for the unrestricted sample. With the
exception of the insignificant flight-3 coefficient, the pattern is for flights 2 through 6 to have
longer buffers than flight 1 (the default), while flights 7 and 8 have shorter buffers than flight
1. Because of its focus on just two flights, the theoretical model does not generate predictions
for the cases covered by the unrestricted sample. But the empirical pattern appears to show
a somewhat different logic than in the two-flight situation. In particular, delay propagation is
apparently addressed through longer buffers for earlier, rather than later flights. The shortest
27 The time-of-day dummies are not used in conjunction with either of the other two variables because they are
almost functionally related. For example, while inclusion of both the time-of-day dummies and flight-sequence
dummies would imply that independent variation is possible for these covariates, the fact that, say, the eighth
flight of the day could never be a morning flight means that such independence is not present. For the same
reason, we do not include the time-of-day dummies in the regressions in columns (1) and (2), as explained
earlier.

30
buffers are for last few flights of the day, given that the aircraft will soon terminate its daily
operations, removing any concern about delay propagation.
The regression in column (7) shows that this buffer pattern also emerges when the flight
dummy variables are replaced by the continous aircraft-rotation variable, which appears in
quadratic form. The positive sign of the aircraft rotation coefficient together with the negative
coefficient on the quadratic term point towards an inverse U-shape relationship between the
flight buffer and the rotation variable, as shown in Figure 6. The non-linear effect of aircraft
rotations on the flight buffer appears to reconcile two opposite forces: the first force pushes
towards longer flight buffers to lessen the risk of delay propagation; the second force pushes
towards shorter flight buffers to maximize aircraft utilization.
Finally, the regressions in Table 6 include a variable measuring on-time performance in the
previous year, on the belief that carriers would change their buffers to remedy past late arrivals.
Using the DOT’s 2017 ‘Reporting Carrier On-Time Performance (1987-present)’ dataset, we
calculated by carrier-route-month the proportion of delayed flights.28 The coefficient on this
past-year delay variable is positive and statistically significant, confirming the expectation that
carriers lengthen flight buffers if delays on the route in the past year were more frequent.

7.5. Ground-buffer regressions with variability proxies

Table 7 shows ground-buffer regressions with variability proxies replacing the variability
measure, doing so for both the two-flight and unrestricted samples. It is worth noting that
the ground buffer is calculated at a flight’s departing airport. In the two-flight sample, the
observed airport corresponds to the departing airport of flight 2, whereas in the unrestricted
sample, the observed airport is the departing airport of Flight i, with i = 2, ..., 8. Thus, the
flight-specific variables in the regression, such as departure time, refer to the following flight.
In this way, the first flight of the day is not included in the regressions for the unrestricted
sample.

28 The ‘Reporting Carrier On-Time Performance (1987-present)’ dataset is essentially the same as the ‘Mar-
keting Carrier On-Time Performance (Beginning January 2018)’, which we use in our empirical analysis, but it
does not report the actual operating carrier. Thus, it does not distinguish between major airlines and affiliated
regional carriers. For this reason and also because an airline could stop serving the route from one year to the
next, some 2018 observations are not matched with the 2017 data.

31
 The effect of weather on ground buffers, as captured by the month dummies, is not as
clear as in the flight-buffer regressions. The results show that in both samples, ground buffers
are longest (relative to January) in the Fall months of September, October and November, a
pattern that does not have a clear weather-based interpretation. However, in the unrestricted
sample, the buffers over the March-August period are significantly shorter than in January,
which is partly consistent with short ground buffers being scheduled in months with better
weather.
The coefficients on the connecting-share and hub-turnaround variables are uniformly pos-
itive and significant across the regressions, as in Table 4. A congested turnaround airport has
longer taxi times, which may prompt the airline to cut ground time, as shown by the nega-
tive coefficients on the congestion-turnaround variable. But adding the slot-control dummy
reverses this effect in both samples, leading to a significantly positive congestion turnaround
coefficient in columns (3) and (6). The significantly negative slot-control effect in the two-flight
subsample is large: ground buffers are almost 3 minutes shorter on average at such airports,
which tend to be congested.29
The pattern of the airline-dummy coefficients in the two-flight sample is somewhat different
than in Table 4, with the coefficients of Delta and Frontier switching to significantly negative
(Southwest’s and Spirit’s coefficients also become more negative). The coefficient pattern
changes somewhat in the unrestricted sample, as seen in columns (4) and (6). The coefficients of
the low-cost and regional-carrier dummy variables, however, follow the same pattern as in Table
4, being negative across regressions. The heavy-aircraft coefficient, previously insignificant, is
now unstable in the two flight sample, but significantly negative in the unrestricted sample.
As in Table 4, the estimated coefficients of the time-of-day dummies show that airlines keep
lengthening ground buffers across the day. This effect is strictly monotonic in each regression,
rising from 2-7 minutes extra buffer in the morning (relative to early morning) to 8-20 minutes
extra buffer in the evening. With the model portraying operation of only two flights, it cannot
predict this pattern. But the pattern suggests that, when an aircraft operates many flights

29 In order to obtain these estimates, we remove the turnaround-airport fixed effects, which are perfectly
collinear with the Slot-controlled airport variable.

32
per day, ground buffers may play a more prominent role than flight buffers in mitigating delay
propagation late in the day.
Airlines may operate with spare capacity during weekends, since most business travel is
mid-week, thus explaining why they set slightly longer ground buffers during weekends, about
two minutes longer on average. Our data show that flights scheduled during weekends are
more punctual than non-weekend fights, and these longer ground buffers may be part of the
reason.30

8. Conclusion
This paper has presented an extensive theoretical and empirical analysis of the choice of
schedule buffers by airlines. With airline delays a continuing problem around the world, such
an undertaking is valuable, and its lessons extend to other transport sectors such as rail and
intercity bus service. One useful lesson from the theoretical analysis of a two-flight model is
that the mitigation of delay propagation is done entirely by the ground buffer and the second
flight’s buffer. The first flight’s buffer plays no role because the ground buffer is a perfect, while
nondistorting, substitute. In addition, the apportionment of mitigation responsibility between
the ground buffer and the flight buffer of flight 2 is shown to depend on the relationship between
the costs of ground- and flight-buffer time.
The empirical results show the connection between buffer magnitudes and a host of vari-
ables, including the month of operation and distance of a flight, whether the flight operates
early or late in the day, and congestion measures at the endpoints. In addition, the initial
regressions relate buffer magnitudes to the variability of flight times, which simulations of
the model identify as an important determining factor (the results show that high variability
lengthens flight buffers).
Fruitful extensions to this work would most likely lie in the theoretical area. The model
could be extended to include additional sequential flights, and the sketch of connecting traffic
provided in the paper could be expanded into a full analysis. In addition, passenger schedul-
ing preferences could be introduced, with a buffer-related extension of scheduled flight times
30 While the average arrival delay of a weekend flight is 6 minutes in the unrestricted sample and 0 minutes
in the two-flight sample, the delays increase to 10 and 3 minutes, respectively, for non-weekend flights.

33
possibly becoming less desirable if it leads to a divergence between a passenger’s preferred and
actual arrival times. The resulting models would be complex, but additional insights could be
generated, with relevance extending beyond the airline industry.
Another avenue of exploration could be in the area of market structure. For example, if a
profit-maximizing airport sets the gate rental cost at an excessive level, the resulting decrease
in ground buffers will impair airline mitigation of delay propagation, with negative effects on
passengers. Alternatively, the entry barrier of gate shortages resulting from airport dominance
by a large carrier (analyzed by Ciliberto and Williams (2010)) would also affect the ability of
smaller carriers to address delay propagation via adequate ground time. Exploration of these
issues could be illuminating.

34
 Appendix

A1. The derivatives of Ω

This appendix section computes the derivatives of the objective function with respect to
bg , b1 , and b2 . Since (2) does not involve bg , the objective function’s derivative with respect to
bg is found by differentiating the sum of (16) and (17). The derivative of the first line of (16)
with respect to bg equals
Z 2
xf1 (b1 + bg ) (2 − b2 )f2 (2 )d2 . (a1)
b2

The second line of (16) can be written as

Z "Z #
1 2
x [1 + 2 − (b1 + b2 + bg )]f2 (2 )d2 f1 (1 )d1
1 =b1 +bg 2 =b1 +b2 +bg −1

Z 1
= x Q(1 , bg )f1 (1)d1 , (a2)
1 =b1 +bg

where Q(1 , bg ) denotes the term in brackets in the first line of (a2). The bg -derivative of (a2)
is
Z 1
∂Q(1, bg )
x f1 (1 )d1 − xQ(b1 + bg , bg )f1 (b1 + bg ). (a3)
1 =b1 +bg ∂bg

Substituting 1 = b1 + bg in the bracketed term in (a2) to evaluate Q(b1 + bg , bg ) in (a3), the

second term in (a3) equals the negative of (a1), so that these terms cancel. The bg -derivative
of (16) is then equal to the first term in (a3).
∂Q/∂bg consists of two components, the first of which comes from differentiating the brack-
eted expression in (a2) with respect to the limit of integration, a derivative that equals zero
upon substituting the limit into the integrand. The second component comes from differenti-
ating with respect to bg under the integral, which yields the bracketed expression in (a2) with
the integrand replaced by −f2 (2 ). Therefore, the first-term in (a3) reduces to

Z "Z #
1 2
−x f2 (2 )d2 f1 (1 )d1 . (a4)
1 =b1 +bg 2 =b1 +b2 +bg −1

35
 Applying the same steps to (17), the bg -derivative of that expression equals

Z "Z #
1 b1 +b2 +bg −1
y f2 (2 )d2 f1 (1)d1 . (a5)
1 =b1 +bg 2 =0

Replacing the bracketed terms in (a4) and (a5) with 1 − F2 (b1 + b2 + bg − 1 ) and F2 (b1 + b2 +
bg − 1 ), respectively, the sum of (a4) and (a5) can be written as

Z 1
= [−x(1 − F2(b1 + b2 + bg − 1 )) + yF2(b1 + b2 + bg − 1 )]f1 (1 )d1
1 =b1 +bg
Z 1  
x ∂Ω
= (x + y) F2(b1 + b2 + bg − 1 )) − f1 (1 )d1 = . (a6)
1 =b1 +bg x+y ∂bg

The derivative of the objective function with respect to b1 builds on the previous results.
The b1 -derivatives of (16) and (17) are identical to the bg derivatives, given by (a4) and (a5),
with their sum equal to (a6). Since (2) also depends on b1 , the b1 -derivative of the objective
function is then the expression in (3) (slightly rearranged) plus (a6):

 
∂Ω x ∂Ω
= (x + y) F1(b1 ) − + . (a7)
∂b1 x+y ∂bg

Using similar steps,

 
∂Ω x ∂Ω
= (x + y)F1(b1 + bg ) F2 (b2 ) − + . (a8)
∂b2 x+y ∂bg

A2. Convexity of Ω
This appendix section first shows that the objective function Ω is strictly convex in b1 and
b2 , with bg fixed at the optimal value, doing so for both cases of zero and positive buffer costs.
The first step is to compute the second derivatives of Ω with respect to b1 and b2 , assuming
zero buffer costs. Using the shorthand Ωij for ∂ 2 Ω/∂bi ∂bj , differentiation of (20) and (21)

36
yields (after suppressing the multiplicative factor x + y))

 
x
Ω11 = f1 (b1 ) − f1 (b1 + bg ) F2(b2 ) −
x+y
Z 1
+ f1 (1 )f2 (b1 + b2 + bg − 1 )d1 (a9)
1 =b1 +bg
Z 1
Ω22 = F1(b1 + bg )f2 (b2 ) + f1 (1 )f2 (b1 + b2 + bg − 1 )d1 (a10)
1 =b1 +bg
Z 1
Ω12 = f1 (1 )f2 (b1 + b2 + bg − 1 )d1 < Ω11 , Ω22 . (a11)
1 =b1 +bg

Observe that the same expressions apply when buffer costs are positive since they vanish in
computing the second derivatives.
Ω22 and Ω12 are positive, while the sign of Ω11 in (a9) is unclear. However, at the optimum
in the zero-cost buffer case, the bracketed term in (a9) is zero, making Ω11 positive and also
ensuring satisfaction of the inequality in (11). The same conclusion holds in the positive-cost
buffer case if cf > cg , in which case the bracketed term in (a9) is negative at the optimum.
With Ω11 , Ω22 > 0 and H ≡ Ω11 Ω22 − Ω212 > 0 (a consequence of the inequalities in (a11)),
Ω is thus strictly convex in b1 and b2 at the optimum. As a result, b1 and b2 values satisfying
the first-order conditions yield a local minimum for Ω, holding bg fixed at its optimal value.
Convexity of Ω in all three buffers cannot be established analytically and must be assumed.

37
Figure 1: Incidence of Delay Propagation

38
 Figure 2: Effect of flight 1's time variability on buffers
1.8

1.6

1.4

1.2

1.0
Buffers

0.8

0.6

0.4

0.2

0.0

-0.2
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.1 1.2 1.3 1.4 1.5
Flight 1's time variability

b1 b2 bg

Figure 3: Effect of flight 2's time variability on buffers

1.8

1.6

1.4

1.2

1
Buffers

0.8

0.6

0.4

0.2

0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.1 1.2 1.3 1.4 1.5
Flight 2's time variability

b1 b2 bg

39
 Figure 4: Effect of common flight time variability on buffers
1.8

1.6

1.4

1.2

1
Buffers

0.8

0.6

0.4

0.2

0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.1 1.2 1.3 1.4 1.5
Common flight time variability

b1,b2 bg

Figure 5: Effect of common flight time variability on buffers

(higher cg)
2

1.5

1
Buffers

0.5

0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.1 1.2 1.3 1.4 1.5

-0.5
Common flight time variability

b1,b2 bg

40
 Figure 6: Flight buffer as a function of aircraft rotation
0.5
0.4
0.3
0.2
Flight buffer

0.1
0
-0.1
-0.2
-0.3
-0.4
-0.5
2 3 4 5 6 7 8
Aircraft rotation

41
 Table 1: Flight 2’s arrival time

Flight 2’s Occurs when Flight 2 late (early) Arrival delay Delay propagation?
departure
On time 1 ≤ b1 + bg as 2 > (≤) b2 max{0, 2 − b2 } NO
Delayed 1 > b1 + bg as 2 > (≤) b1 + b2 + bg − 1 max{0, 1 + 2 − (b1 + b2 + bg )} YES if late

42
 Table 2: Description and main statistics of the empirical variables
b bg Variables Description Two-flight Unrestricted
sub-sample sample
X Flight buffer (b) Difference between the scheduled flight time and 39.95 31.73
the minimum actual flight time, computed by (15.87) (12.89)
route and aircraft type
X Ground buffer (bg ) Difference between the scheduled ground time 40.30 37.88
and minimum feasible ground time, computed by (30.46) (25.96)
turnaround airport and aircraft type
X X Flight 1’s variability Flight 1’s standard deviation of flight-time 14.31
(6.53)
X X Flight 2’s variability Flight 2’s standard deviation of flight-time 14.54
(7.01)
X X Connecting passengers Share of past-year connecting passengers, proxy 0.31 0.45
based on DB1B Market dataset (0.24) (0.28)
X X February-December Set of monthly dummy variables, January is the
omitted month
X Hub origin Dummy variable = 1 if airport of origin is the hub 0.51 0.36
of the airline (0.50) (0.48)
X Hub destination Dummy variable = 1 if airport of destination is the 0.51 0.36
hub of the airline (0.50) (0.48)
X Hub turnaround Dummy variable = 1 if airport of turnaround is 0.45 0.38
the hub of the airline (0.49) (0.49)
X Congestion origin Number of landing and departing flights at the air- 13.12 11.64
port of origin on the same hour when the flight is (7.08) (7.64)
scheduled to depart divided by the number of run-
ways of the airport of origin
X Congestion destination Number of landing/departing flights at the airport 12.23 11.39
of destination on the same hour when the flight is (6.81) (7.72)
scheduled to land divided by the number of run-
ways of the airport of destination
X Congestion turnaround Number of landing and departing flights at the air- 12.42 12.42
port of turnaround on the same hour when incom- (7.14) (7.72)
ing flight is scheduled to land divided by the num-
ber of runways of the airport of turnaround
X Slot controlled airport Dummy variable = 1 if airport of turnaround is 0.07 0.05
slot controlled (DCA, JFC and LGA) (0.23) (0.22)
X X Alaska-Southwest Set of airline dummy variables, American Airlines
is the omitted airline
X X Regional carrier Dummy variable = 1 if the flight is operated by a 0.07 0 .31
regional carrier (0.26) (0.47)
X Low-cost carrier Dummy variable = 1 if the flight is operated by 0.16 0.29
a low-cost carrier (i.e. Jet Blue, Frontier Airlines, (0.46) (0.45)
Allegiant Air, Spirit Airlines and Southwest Air-
lines)
X Competitors Number of competitors on the route 1.58 1.06
(1.21) (1.11)
X Distance Route distance, in 100-mile units 16.94 8.21
(8.66) (5.92)

continuing next page

43
Table 2: Description and main statistics of the empirical variables (continued)
b bg Variables Description Two-flight Unrestricted
sub-sample sample
X Past-year delay Share of past-year delayed flights, computed by 0.20 0.17
carrier-route-month (0.11) (0.11)
X X Morning-Evening Set of departure time variables, Morning (9.00-
11.59), Afternoon (12.00-15.59), Late afternoon
(16.00-17.59) and Evening (18.00-23.59); the omit-
ted category is Early morning (0.00-8.59)
X X Weekend Dummy variable = 1 if the flight departs in the 0.31 0.27
weekend (0.46) (0.44)
X Flight i Dummy = 1 for the ith flight operated in a day
by a given aircraft, uniquely identified by its tail
number
X Aircraft rotation Sequence of flights operated in a day by a given 2.99
aircraft (1.69)
X X Heavy aircraft Dummy variable = 1 the if aircraft has been as- 0.23 0.03
signed a maximum takeoff weight rating of 300,000 (0.42) (0.17)
lb or more
X Managerial origin Managerial share of the origin city’s work-force 5.69
(1.30)
X Managerial turnaround Managerial share of the turnaround city’s work- 5.65
force (1.28)
(a) The symbol Xdenotes whether the variable is included in the flight buffer regressions (column b) or in the ground buffer
regressions (column bg ).
(b) The second-last and last columns respectively report the mean value of the variable for the two-flight sub-sample and
unrestricted sample with the standard errors in parentheses. Empty cells on some variables are due to space reason, since their
inclusion would require reporting a set multiple dummies whose mean values would not be very informative.

44
 Table 3: Flight-buffer regressions with flight-time variability
(1) (2) (3) (4)
Dependent variable b1 b2 b1 , b2 b1 , b2
Flight 1’s variability 0.556*** 0.200*** 0.377*** 0.375***
Flight 2’s variability 0.104*** 0.385*** 0.240*** 0.243***
Connecting pax -17.000*** -17.814*** -17.698*** -17.677***
Hub origin 2.198*** 2.207*** 2.237*** 2.113***
Hub destination 2.097*** 3.630*** 2.892*** 2.959***
Alaska Airlines -0.849 1.372** 0.085 0.119
Allegiant Air -14.970*** -11.735*** -13.615*** -13.481***
Delta Airlines 1.781*** 2.080*** 1.899*** 1.916***
Frontier Airlines -10.570*** -10.563*** -10.534*** -10.565***
Hawaiian Airlines -0.802 -4.120*** -2.099*** -2.224***
Jet Blue -3.490*** -4.081*** -3.902*** -3.920***
Southwest Airlines -6.038*** -6.016*** -6.014*** -5.945***
Spirit Airlines -16.030*** -13.608*** -14.927*** -14.960***
United Airlines -0.792 -1.570*** -1.131*** -1.161***
Virgin America 2.885 6.003*** 4.144** 4.230**
Regional carrier -8.215*** -6.834*** -7.358*** -7.371***
Competitors 1.108*** 1.312*** 1.216*** 1.217***
Managerial origin 1.370*** 1.386*** 1.367*** 1.402***
Heavy aircraft 4.038*** 2.696*** 3.399*** 3.392***
Morning 1.031***
Afternoon -0.762***
Late Afternoon 0.742**
Evening -0.167
Flight 2 -0.829***
Constant 24.907*** 24.357*** 24.810*** 25.126***
R2 0.306 0.354 0.319 0.318
Observations 113,873 104,502 218,375 218,375
(a) The estimated coefficients marked with ***, ** and * are statistical significance at, respectively the 1%, 5% and 10% level.
(b) The standard errors, not reported to save space, are clustered by route-month.

45
 Table 4: Ground-buffer regressions with flight-time variability
(1) (2) (3) (4)
Dependent variable bg bg bg bg
Flight 1’s variability -0.034 -0.028 0.242*** 0.209***
Flight 2’s variability -0.091*** -0.118*** -0.178*** -0.302***
Connecting pax 10.709*** 9.520*** 11.600*** 9.728***
Hub turnaround 13.201*** 12.698*** 13.106*** 12.642***
Alaska Airlines 8.258*** 8.162***
Allegiant Air -13.489*** -12.382***
Delta Airlines 2.112*** 1.869***
Frontier Airlines 4.403** 4.764**
Hawaiian Airlines 25.918*** 25.718***
Jet Blue -3.248*** -3.878***
Southwest Airlines -1.929** -1.288
Spirit Airlines -2.438*** -2.412***
United Airlines 9.995*** 9.900***
Virgin America 10.426*** 10.452***
Regional carrier -1.070** -0.539
Low-cost carrier -8.064*** -8.485***
Managerial turnaround 0.538*** 0.900*** 0.519*** 0.909***
Heavy aircraft -0.593 0.453 -1.010 0.058
Morning 4.546*** 4.112*** 4.127*** 3.640***
Afternoon 11.897*** 13.096*** 11.335*** 12.520***
Late Afternoon 12.587*** 13.328*** 12.166*** 12.938***
Evening 14.591*** 15.493*** 13.997*** 14.862***
Constant 15.275*** 18.104*** 13.027*** 18.250***
2
R 0.135 0.109 0.136 0.111
Observations 124,131 124,131 124,131 124,131
(a) The estimated coefficients marked with ***, ** and * are statistical significance at, respectively the 1%, 5% and 10% level.
(b) The standard errors, not reported to save space, are clustered by route-month.
(c) Flight-time variability calculated by route, month and flight i in columns (1) and (2), by route and flight i in columns (3)
and (4),

46
 Table 5: The determinants of flight-time variability
(1) (2)
Dependent variable Flight 1’s Flight 2’s
variability variability
February -0.558** -0.872***
March -1.080*** -1.205***
April 0.604** -0.503*
May -0.297 0.183
June -0.380 0.128
July 1.134*** 0.921***
August 1.104*** 1.633***
September -0.681** -0.115
October -0.968*** -0.913***
November -0.420 0.180
December 0.107 0.023
Congestion origin 0.025** 0.020
Congestion destination 0.036** 0.064***
Distance 0.135*** 0.152***
Weekend -0.109** -0.119***
Constant 5.957* -8.461*
R2 0.283 0.281
Observations 114,011 113,195
(a) The estimated coefficients marked with ***, ** and * are statistical significance at, respectively the 1%, 5% and 10% level.
(b) The standard errors, not reported to save space, are clustered by route-month.
(c) All estimates include airport of origin and airport of destination fixed effects.

47
 Table 6: Flight-buffer regressions with variability proxies
(1) (2) (3) (4) (5) (6) (7)
Dependent variable b1 b2 b1 , b2 b1 , b2 b1 , ..., b8 b1 , ..., b8 b1 , ..., b8
Two-flight sub-sample Unrestricted sample
February 0.740 0.283 0.548 0.547 0.161 0.164 0.164
March -0.762 -0.645 -0.717 -0.706 -1.679*** -1.663*** -1.662***
April -4.153*** -0.900* -2.563*** -2.563*** -2.930*** -2.915*** -2.914***
May -6.551*** -0.248 -3.564*** -3.560*** -3.756*** -3.742*** -3.742***
June -8.256*** -0.413 -4.516*** -4.513*** -4.369*** -4.347*** -4.346***
July -8.743*** -0.727 -4.861*** -4.865*** -4.628*** -4.608*** -4.607***
August -8.977*** -1.279** -5.192*** -5.202*** -4.589*** -4.572*** -4.571***
September -8.970*** -3.034*** -5.914*** -5.943*** -4.646*** -4.638*** -4.637***
October -8.501*** -3.735*** -6.041*** -6.062*** -4.521*** -4.509*** -4.508***
November -5.134*** -3.523*** -4.178*** -4.198*** -2.427*** -2.417*** -2.416***
December -2.819*** -2.874*** -2.830*** -2.827*** -1.338*** -1.332*** -1.332***
Connecting passengers -0.636 0.300 -0.050 0.026 0.182 0.194 0.199
Hub origin 1.540*** -0.034 0.037 0.108 0.039 0.012 0.034
Hub destination 1.946*** 2.691*** 2.673*** 2.691*** 1.026*** 1.041*** 1.024***
Congestion origin 0.155*** 0.175*** 0.130*** 0.127*** 0.195*** 0.188*** 0.188***
Congestion destination 0.085*** 0.131*** 0.136*** 0.134*** 0.141*** 0.133*** 0.132***
Alaska Airlines 1.404** 1.045 2.415*** 2.490*** 0.063 0.076 0.075
Delta Airlines -0.286 -0.866 -0.599 -0.583 0.483*** 0.473*** 0.477***
Frontier Airlines -1.635 -1.768 -1.604* -1.260 -0.582*** -0.574*** -0.554***
Hawaiian Airlines -1.707* -6.627*** -4.348*** -4.371*** -7.235*** -7.328*** -7.310***
Jet Blue -4.847*** -4.617*** -5.532*** -5.319*** -3.807*** -3.789*** -3.781***
Southwest Airlines -1.391** -2.118*** -1.797*** -1.697*** 0.846*** 0.859*** 0.862***
Spirit Airlines -4.294*** -4.178*** -4.318*** -4.102*** -4.104*** -4.105*** -4.088***
United Airlines -3.932*** -5.364*** -4.702*** -4.666*** -2.307*** -2.301*** -2.305***
Virgin America -9.026*** -5.956*** -7.990*** -7.889*** -5.518*** -5.506*** -5.502***
Regional carrier -0.735* -0.070 -0.108 -0.143 2.208*** 2.186*** 2.194***
Competitors -0.173 0.709*** 0.284* 0.276* 0.211*** 0.207*** 0.205***
Distance 1.150*** 1.202*** 1.189*** 1.186*** 1.457*** 1.457*** 1.456***
Heavy aircraft -2.055*** -1.681*** -1.790*** -1.829*** -4.852*** -4.837*** -4.833***
Past-year delay 11.388*** 5.230*** 9.171*** 9.218*** 3.978*** 4.011*** 4.011***
Weekend -0.080 -0.110 -0.090 -0.093* 0.066*** 0.044*** 0.044***
Morning 0.186 -0.140***
Afternoon 0.845*** -0.349***
Late Afternoon 1.278*** 0.557***
Evening 0.873*** 0.152***
Flight 2 0.883*** 0.178***
Flight 3 -0.039
Flight 4 0.227***
Flight 5 0.253***
Flight 6 0.125**
Flight 7 -0.610***
Flight 8 -1.052***
Aircraft rotation 0.238***
Aircraft rotation2 -0.035***
Constant 29.483*** 16.645*** 19.836*** 20.015*** 14.388*** 14.436*** 14.177***
R2 0.676 0.680 0.658 0.658 0.654 0.654 0.654
Observations 101,920 100,916 202,836 202,836 4,126,741 4,126,741 4,126,741
(a) The estimated coefficients marked with ***, ** and * are statistical significance at, respectively the 1%, 5% and 10% level.
(b) The standard errors, not reported to save space, are clustered by route-month.
(c) All estimates include airport of origin and airport of destination fixed effects.

48
 Table 7: Ground-buffer regressions with variability proxies
(1) (2) (3) (4) (5) (6)
Dependent variable bg bg bg bg bg bg
Two-flight sub-sample Unrestricted sample
February 0.246 0.271 0.056 -0.184 -0.176 -0.316
March -0.773 -0.791 -1.284 -1.082*** -1.092*** -1.455***
April -0.472 -0.716 -1.305 -0.862*** -0.871*** -1.210***
May 1.036 0.778 0.529 -0.159 -0.170 -0.486**
June -0.629 -0.912 -1.224 -1.224*** -1.234*** -1.767***
July 0.475 0.139 -0.048 -0.906*** -0.909*** -1.467***
August 1.317 0.993 0.754 -0.563*** -0.578*** -1.170***
September 3.224*** 2.935*** 2.773*** 1.274*** 1.315*** 0.829***
October 2.814*** 2.543*** 2.102** 0.893*** 0.868*** 0.360*
November 1.865** 1.705** 1.198 0.917*** 0.921*** 0.457**
December 0.382 0.205 0.015 0.433** 0.404** 0.063
Connecting passengers 8.245*** 9.167*** 7.904*** 4.499*** 2.172*** 1.229***
Hub turnaround 14.250*** 13.050*** 10.085*** 14.013*** 13.477*** 14.025***
Congestion turnaround -0.375*** -0.368*** 0.072** -0.388*** -0.396*** 0.153***
Slot controlled airport -2.714*** -0.335*
Alaska Airlines 1.578* 5.114*** 6.646*** 8.595***
Allegiant Air -11.395*** -21.571*** -5.952*** -10.330***
Delta Airlines -3.080*** -0.664 -3.060*** -1.160***
Frontier Airlines -4.015** -1.468 -1.167*** -0.390
Hawaiian Airlines -4.920*** 7.824*** -4.987*** 8.641***
Jet Blue -3.958*** -6.371*** 0.515 0.089
Southwest Airlines -6.813*** -7.743*** -7.012*** -5.465***
Spirit Airlines -6.037*** -6.573*** 1.925*** 3.299***
United Airlines 4.644*** 7.802*** 4.924*** 9.248***
Virgin America 7.170*** 6.824*** 9.104*** 11.603***
Regional carrier -1.190* -6.839*** -1.740*** -4.471***
Low-cost carrier -5.649*** -4.996***
Heavy aircraft -0.694 -1.121* 1.503** -2.692*** -3.721*** -1.027***
Morning 7.225*** 7.033*** 5.976*** 4.205*** 4.240*** 1.979***
Afternoon 14.282*** 14.122*** 14.192*** 4.453*** 4.396*** 2.816***
Late Afternoon 19.239*** 19.341*** 16.752*** 5.840*** 5.915*** 3.907***
Evening 19.951*** 20.043*** 19.082*** 9.379*** 9.575*** 8.089***
Weekend 1.932*** 1.906*** 2.647*** 1.867*** 1.918*** 2.361***
Constant 27.496*** 24.564*** 37.866*** 23.269*** 22.581*** 32.451***
R2 0.206 0.201 0.157 0.237 0.227 0.202
Observations 133,178 133,178 133,178 4,318,387 4,318,387 4,318,387
(a) The estimated coefficients marked with ***, ** and * are statistical significance at, respectively the 1%, 5% and 10% level.
(b) The standard errors, not reported to save space, are clustered by route-month.
(c) All estimates but columns (3) and (6) include airport of turnaround and airport of origin fixed effects; columns (3) and (6)
include only airport of origin fixed effects.

49
 References

AhmadBeygi, S., Cohn, A., Lapp, M., 2010. Decreasing airline delay propagation by
re-allocating scheduled slack, IIE Transactions 42, 478-489.

Arikan, M., Deshpande, V., Sohoni, M., 2013. Building reliable air-travel infrastructure
using empirical data and stochastic models of airline networks. Operations Research 61,
45-64.

Ball, M., Barnhart, C., Dresner, M., Hansen, M., Neels, K., Odoni, A., Pe-
terson, E., Sherry, E., Trani, A., Zou, B., 2010. Total Delay Impact Study: A
Comprehensive Assessment of the Costs and Impacts of Flight Delay in the United States.
Final Report, National Center of Excellence for Aviation Operations Research (NEXTOR).

Barnhart, C., Cohn, A., 2004. Airline schedule planning: Accomplishments and opportu-
nities. Manufacturing & Service Operations Management 6, 3-22.

Britto, R., Dresner, M., Voltes, A., 2012. The impact of flight delays on passenger
demand and societal welfare. Transportation Research Part E 48, 460-469.

Brueckner, J.K., Czerny, A.I., Gaggero, A.A., 2020. Airline schedule buffers and
flight delays: A discrete model. Unpublished paper, UC Irvine.

Ciliberto, F., Williams, J.W., 2010. Limited access to airport facilities and market power
in the airline industry. Journal of Law and Economics 53, 467-495.

Deshpande, V., Arikan, M., 2012. The impact of airline flight schedules on flight delays.
Manufacturing & Service Operations Management 14, 423-440.

Forbes, S.J., 2008. The effect of air traffic delays on airline prices. International Journal of
Industrial Organization 26, 1218-1232.

Forbes, S.J., Lederman, M., 2010. The performance implications of vertical integration:
Evidence from the airline industry. RAND Journal of Economics 41, 765-790.

Forbes, S.J., Lederman, M., 2009. Adaptation and vertical integration in the airline
industry. American Economic Review 99, 1831-49.

Forbes, S.J., Lederman, M., Wither, M.J., 2019. Quality disclosure when firms set
their own quality targets. International Journal of Industrial Organization 62, 228-250.

Forbes, S.J., Lederman, M., Yuan, Z., 2019. Do airlines pad their schedules? Review of

50
 Industrial Organization 54, 61-82.

Hao, L., Hansen, M., 2014. Block time reliability and scheduled block time setting. Trans-
portation Research Part B 69, 98-111.

He, W., 2019. Integrating overbooking with capacity planning: Static model and application
to airlines. Production and Operations Management 28, 1972-1989.

Kafle, N., Zou, B., 2016. Modeling flight delay propagation: A new analytical-econometric
approach. Transportation Research Part B 93, 52-42.

Kang, L., Hansen, M., 2017. Behavioral analysis of airline scheduled block time adjustment.
Transportation Research Part E 103, 56-68.

Kang, L., Hansen, M., 2018. Assessing the impact of tactical airport surface operations on
airline schedule block time setting.Transportation Research Part C 89, 133-147.

Luttmann, A., 2019. Evidence of directional price discrimination in the U.S. airline industry.
International Journal of Industrial Organization 62, 291-329.

Mayer, C., Sinai, T., 2003. Why do airlines systematically schedule their flights to arrive
late? Unpublished paper, Wharton School of Business, University of Pennsylvania.

Mazzeo, M., 2003. Competition and service quality in the U.S. airline industry. Review of
Industrial Organization 22, 275-296.

Prince, J.T., Simon, D.H., 2015. Do incumbents improve service quality in response to
entry? Evidence from airlines on-time performance. Management Science 61, 372-390.

Shumsky, R.A., 1993. Response of U.S. air carriers to on-time disclosure rule. Transportation
Research Record 1379, 9-16.

Sohoni, M., Lee, Y.-C., Klabjan, D., 2011. Robust airline scheduling under block-time
uncertainty. Transportation Science 45, 451-464.

Whitin, T.M., 1955. Inventory control and price theory. Management Science 2, 61-68.

Wang, X., 2015. Service competition in the airline industry: Schedule robustness and market
structure. Unpublished paper.

Wang, Y., Zhou, Y., Hansen, M., Chin, C., 2019. Scheduled block time setting and
on-time performance of U.S. and Chinese airlines: A comparative analysis. Transportation
Research Part A 130, 825-843.

51
Zhang, A., Czerny, A.I., 2012. Airports and airlines economics: An interpretive review of
recent research. Economics of Transportation 1, 15-34.

Zhang, D.J., Salant, Y., Van Mieghem, J.A., 2018. Where did the time go? On the
increase in airline schedule padding over 21 years. Unpublished paper.

Zou, B., Hansen, M., 2012. Impact of operational performance on air carrier cost structure:
Evidence from US airlines. Transportation Research Part E 48, 1032-1048.

52