Commercial Aircraft Departure Delays

Rebecca Nugent

rnugent@[Link]

Introduction:

Humanity is increasingly global and connected, and airline travel is an indespensible part of this modern connectivity.
Airline flight delays have thus been a concern for a number of years, even before Covid-related issues [1]. Airline departure
delays are not only frustrating for travelers but also costly for travelers, airlines, and the overall economy [2]. Our goal in the
current work is therefore to better understand airline departure delays. This work may be of help to airlines in improving
overall adherence to their flight schedules.

Data:

The data contain information on a sample of 500 domestic commercial airline flights from 2008, collected by the U.S.
Bureau of Transportation Statistics [3].

For each flight, the data records the month of the flight, the day of the week of the flight, the total air time (in minutes),
the total distance (in miles) the plane travelled, the minutes it took the plane to taxi out on the runway at departure, the
minutes it took to taxi in on the runway at destination, the plane’s departure delay (in minutes) [could be zero if the plane
departed on time, or negative if the plane departed early],the arrival delay (in minutes) at destination [could be zero if the
plane arrived on time, or negative if the plane arrived early], whether the plane was diverted (yes/no), whether there was a
delay due to something involving the airline carrier (yes/no), whether there was a delay due to the weather (yes/no),
whether there was a delay due to the National Air System (yes/no), whether there was a delay due to security, and whether
there was a delay due to late aircraft (yes/no). There are no missing values in any of the variables in the dataset.

Of primary focus is the variable departure delay, a quantitative variable measuring how much the flight was delayed before
take off (measured in minutes); the mean departure delay was 48.678 minutes, with standard deviation of departure delay
52.68 minutes. The research questions [stated in the next section] will also include day of the week, and arrival delay. The
variable day of the week is a categorical variable indicating which day of the week a flight took place. Friday was the most
popular day for flights in the data (94 of the flights were on Friday, comprising 18.8% of the 500 flights) and Sunday was
the least popular (60 of the flights were on Sunday, comprising 12.0% of the 500 flights). Arrival delay is a quantitative
variable measuring how much the flight was delayed on arrival (measured in minutes); the mean arrival delay was 55.81
minutes, and the standard deviation of arrival delaya was 49.76 minutes.

Research Scenarios:

Characterizing the Distribution of Departure Delay

What is the nature of the departure delays? How long do departure delays tend to be? Are most of them relatively long?
Many frequent fliers report experiencing departure delays of less than half an hour (perhaps the airlines are working to
maintain the flight schedule in order to minimize flight disruption costs and customer dissatisfaction; future survey to verify
this could be a direction for future research), so it is interesting to test this anecdotal claim with the data.

The histogram of departure delay (figure 1) indicates that most of the flights do have a relatively short departure delay (of
between 10 to 20 minutes). The distribution of departure delay shows one mode around 40 to 50 minutes, is strongly right-
skewed (long delays are rare), but has some notable potential high outliers around 250 minutes (over 4 hours), 300 minutes
(5 hours), and 350 minutes (almost 6 hours). Also of note are about 40 flights with zero delays or negative departure delays
(flights that were on time or left early).
 FIGURE 1: Distribution of Departure Delay

The mean departure delay was 48 minutes, with a relatively large standard deviation of almost 53 minutes. The median
departure delay is 33 minutes; note that therefore more than half the flights had departure delay of more than half an hour, so
the anecdotal claim that departure delays are less than half an hour is contradicted by the data. The middle half of departure
delays span from a 25th percentile of 15.5 minutes to a 75th percentile of 65.5 minutes. Since only 25% of the flights have
departure delay below 15.5 minutes, this provides further evidence that most of the flights have non-trivial delays.

Relationship between Departure Delay and Day of the Week

Does airline departure delay vary by day of the week, and if so, what is the particular nature of the relationship? It seems
reasonable to hypothesize that there might be different distributions of departure delays on different days, since there are
likely different types of people flying at different times (e.g., more likely business commuters on Monday morning, versus
more likely tourists on Fridays).

The mean and standard deviation of departure delay by each day of the week is summarized in table 1.

TABLE 1: Departure Delay by Day of Week

Monday Tuesday Wednesday Thursday Friday Saturday Sunday

Mean Delay (min) 48.40 53.52 44.52 46.40 52.93 44.10 49.05

Std Dev Delay (min) 47.88 56.64 47.60 52.22 58.57 53.42 50.86

From the summaries in table 1, some minor variation of departure delay across the days of the week is evident: Tuesday and
Friday seem to have the longest average departure delays where Wednesday and Saturday have slightly shorter delays.

Graphical investigation of the relationship between departure delay and day of week is shown in the side by side boxplot in
figure 2.
 FIGURE 2:
Distribution of Departure Delay by Day of Week

From the boxplots in figure 2, it is notable that the median of departure delay is slightly higher on Tuesday than on other
days (thus showing a similar comparative behavior as the computed means in table 1); however, some notable high outliers
are evident in figure 2, in particular on Friday, so the average delay on Friday might be attributable to the effect of those
outliers. Monday and Saturday also have noticeable large outliers, with the longest delay of 350 minutes occurring on
Saturday (which is probably one of the most popular travel days).

Overall, although there are some small differences in means and medians of departure delay across the days of the week, the
variation is sufficiently large and the boxplots sufficiently similar that there is not much evidence of a relationship between
departure delay and day of the week.

Relationship between Arrival Delay and Departure Delay

Is there a relationship between arrival delay and departure delay for flights? If so, what are the characteristics of the
relationship, and does it depend on the reason for the delay?

It seems reasonable to suspect that generally if a flight leaves late it will arrive late. However, in some cases, a particular
flight that departs late might be able to make up lost time and minimize its arrival delay at its destination; but if the reason
for the delay is an ongoing problem that is out of the control of individual flights (e.g., bad weather), delays may compound
and having a departure delay might actually correspond to a worse arrival delay.

The relationship between arrival delay and departure delay is illustrated in the scatterplot in figure 3. The strong positive
relationship indicates that greater departure delay strongly predicts greater arrival delay. Two flights lie notably somewhat
outside the trend, one with negative departure delay (the flight left early) but arrival delay around 150 minutes, and the other
departing only about 25 minutes late but arriving delayed by almost 200 minutes (almost 4 hours). The flight with the
longest departure delay (noted as a Saturday flight from figure 2) is now seen on the scatterplot to also have the longest
arrival delay.
 FIGURE 3: Arrival Delay vs. Departure Delay

Right-skewness of both Arrival Delay and Departure Delay are evident on figure 3 (seen by the density of flights near zero
departure delay and zero arrival delay, and the increasing sparseness/thinning of flights for larger delays). Aside from that,
however, the relationship looks sufficiently linear that a linear regression model and correlation coefficient are reasonably
appropriate. The correlation between Arrival Delay and Departure Delay is 0.917, so very strong; and the regression
equation is given in equation 1.

EQUATION 1: Linear Regression of Departure Delay on Arrival Delay

Departure Delay = 13.65 + 0.87*Arrival Delay

Note that the slope is slightly less than 1 which indicates that for each additional minute of departure delay there is less than
a minute of additional arrival delay predicted, on average. Hence, based on the regression model, time delays from departure
to arrival do not compound, at least on average.

Flight delays can occur due to problems that can affect the whole flight system (for instance weather, or problems with the
airline’s computer system), or something that only makes the particular flight late (for instance a security problem).
Interestingly, nearly half the flights in the sample (246 flights, 49.2% of the flights) had a delay because of late aircraft. With
this in mind, in figure 4, the relationship between arrival delay and departure delay is labeled and colored by whether or not
the delay was due to late aircraft.
 FIGURE 4: Arrival Delay vs. Departure Delay,
labeled by Late Aircraft Status

In figure 4, the relationship between departure delay and arrival delay is notably similar for flights that had delay due to late
aircraft and those that were delayed for other reasons, with a slightly higher intercept (higher arrival delay, when there was
no departure delay) for the flights without late aircraft. Table 2 catalogs the intercepts and slopes of the corresponding
regression lines, as well as the correlation coefficients.

TABLE 2: Regression and Correlation, for predicting Arrival Delay from Departure Delay, depending on cause (Late
Aircraft or Not Late Aircraft)

Late Aircraft (n = 246) No Late Aircraft (n = 254)

Intercept 4.630 20.346

Slope 0.950 0.790

Correlation coefficient 0.960 0.861

Note that the slopes of the linear models are relatively similar, and both slightly less than 1 which is similar to the slope for
the regression model for the unlabeled data. We note that the intercepts are relatively different.

Discussion:

Departure delays for a sample of 500 flights from 2008 were characterized, and examined for potential relationships with
arrival delays, late aircrafts, and day of the week. Overall, flights have a large range of departure delays with the majority
between 30-60 minutes. There was not a strong relationship with day of the week due to the large variation in delays across
all days. While the relationship between arrival delay and departure delay was strong (as expected), it did not seem to
change much whether or not there was a late aircraft. Future analyses might include season as well; we would expect delays
to be weather-dependent. We might also ask the Bureau for airline carrier info, since it seems plausible there are some
airlines with lengthier average delays than others.
Overall, this analysis can help both passengers and airlines understand, plan for, and possibly mitigate late aircraft flights.
References:

[1] New York Times, Sept. 16, 2019, [Link]

[2] 2010 NEXTOR report, [Link]

[3] Bureau of Transportation Statistics, [Link]