An Econometric Study of Flight Delay Causes at O’Hare International Airport

Nathan Daniel Boettcher
Dr. Don Thompson*
Pepperdine University

Keywords: Econometrics, Flight Delay, O'Hare, Delay Causes

Abstract

This study examined the relationship between sources of delay and the level of delay at Chicago’s O’Hare airport. Measuring delay as an aggregation of minutes wasted, rather than a percentage of total on-time flights, led us to conclude that diminishing individual airlines’ contributions to delay would be ineffective in decreasing the aggregate minutes delayed. Our results suggest that policies aimed at reducing delays due to late arriving aircraft would be the most efficacious in lowering the minutes delayed at O’Hare. This result also proposes that further study is required on how to diminish the contribution of a specific source of delay, as opposed to targeting delay as a concept at large.

Introduction

Flight is a unique transportation phenomenon with both unpredictable and uncontrollable factors. The steady increase of flight volume over the past decades indicates the growing significance of reliable transportation, and has motivated further research to comprehend the various mechanisms that govern on-time performance at major U.S. airports.

As such, the prior research on this topic is plentiful and employs a wide variety of methods, all of which focus on diminishing levels of delay. The FAA itself has published several reports, such as the Airport Capacity Benchmark Report (2004), in which each major airport is individually analyzed and given a range of flights per hour based on condition. This report also makes several recommendations for planning new runways and structures based on their findings. While this study is no doubt helpful in scheduling flights, and determining operating capacity during sub-optimal conditions, it makes little comment on the role of airlines on either level of delay, or the cost of said delay.

Other researchers have studied possible airport demand management strategies (Fan, 2003) to evaluate how airports should handle increased demand on the system, such as the high volume of travelers during the holidays. This paper serves to model a type of policy recommendation aimed at lowering flight delay by changing FAA regulations in regards to managing delay during high demand periods.

There are also several papers which employ a queueing model to determine the effect of airport externalities, such as weather (Allan, et al., 2001) or flight traffic congestion (Hansen, 2002). Other papers have developed predictive models of delay (Evans, 2008) that are analogous to models of queueing theory, yet can be run at much faster speeds than the queuing simulation.

Lastly, a report was released by the American Aviation Institute (2012), which included broad analysis of delay-related phenomena, and found several key conclusions, the most relevant of which is that they found no correlation between airline scheduling practices and the main sources of delay are outside of airline control.

This present work is unique in its focus on the cost to consumers of wasted time as a consequence of delayed flight, instead of the usual emphasis on system level of delay. From the perspective of a consumer, their primary concern is the duration of their delay, not the aggregate level of delay at a given airport. To demonstrate this, consider a day in which 100 out of a possible 1,000 flights are delayed, each by 30 minutes. Now compare this to a day in which 50 out of a possible 1,000 flights are delayed, each by 60 minutes. Using metrics that measure level of delay either by on-time percentage or by number of flights delayed would suggest that the 50 flight delayed day is preferable to the 100 flight delayed day. However, from the perspective of potential consumers, the cost of wasted time is equal: 3,000 wasted minutes for each day. Motivated by this concern, our study sought to analyze how to decrease minutes delayed, as opposed to the percentage of flights delayed. Specifically, we focused on the role of individual airlines’ contribution to minutes delayed by comparing their effect to the other sources of delay. We chose to examine Chicago’s O’Hare International, due to its mediocre on-time performance, flight traffic congestion, and high proportion of connecting flights.

Data

We accessed data from the FAA’s ASPM and ASQM databases. The study chose to analyze Chicago’s O’Hare monthly delay statistics from July 2003 through December 2014.

Methods

We conducted a wide array of statistical analysis. Ultimately, an ordinary least squares time series regression arose as the best method to determine the effect of the various delay sources. We ran a variety of models, and detail the best-fit model below.

Model

First, it is important to note that neither the setup nor interpretation of our model is intuitive. Our model is similar to an investor who runs a forecasting model on the historical performance of various investment options in order to determine the best investment distribution. Keeping that concept in mind, let’s assume that the central planners at O’Hare could choose to distribute the level of delay in their system across various delay causes. For example, if they knew that 100 flights were going to be delayed, suppose that they have the power to assign x flights to be delayed by airline, y flights delayed by late arriving aircraft, and z flights delayed by the national aviation system. Let’s also suppose that some w flights will be delayed due to weather, and that this is out of the central planner’s control. So, x+y+z=100-w flights. Our research focused on these sources of delay since they are the standard causes reported in an FAA delay cause report. This is similar to an investor having $100 to invest, where he/she can choose to invest $x in investment a, $y in investment b, etc. Based on historical performance, the investor could determine an optimal distribution of his/her funds into the various investment opportunities.

In a similar fashion, we wanted to determine how the distribution of flights delayed among various sources affects the total numbers of minutes delayed. Choosing to focus on individual airlines’ contribution to delay, we then used a linear regression, which predicts the percent change in total minutes delayed based on the share of delay due to carrier. There are also variables included to control for delay due to trends over time, weather, and excess demand on the airport’s capacity. This model thus would reveal whether airlines, as a source of delay, tend to increase total minutes delayed.

Results

The model given in Tables 1 and 2 was conducted using Newey-West standard errors to help correct for heteroskedasticity and serial autocorrelation present in the data. The results are summarized below:

Table 1

Table 2

Regression with Newey-West standard errors          Number of obs  =       138
maximum lag: 12                                     F(  5,   132)  =     41.42
R-squared: .676                                     Prob > F       =    0.0000

--------------------------------------------------------------------------------
               |             Newey-West
      logtotmi |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
---------------+----------------------------------------------------------------
       carrier |  -.1057365   .0187025    -5.65   0.000    -.1427318   -.0687411
carriersquared |   .0009968   .0003179     3.14   0.002     .0003679    .0016256
            wx |   .0222324   .0203948     1.09   0.278    -.0181106    .0625754
          time |   .0026153   .0012798     2.04   0.043     .0000838    .0051469
      logschar |   1.258459   .4173042     3.02   0.003     .4329901    2.083928
         _cons |  -1.707978   1.565277    -1.09   0.277     -4.80425    1.388294

- The quadratic modelling with respect to carrier delay implies that the rate of change is non-constant, since the derivative is a linear function. Thus, interpreting the effect of increasing delay due to carrier depends upon the value of carrier delay. To provide a baseline interpretation, we use the mean carrier delay as the interpretation value, 20.53 percent.

- Controlling for weather, time trends, and the number of scheduled arrivals, the mean share of delay due to carrier, 20.53 percent, is associated with a 6.48 percent decrease in total minutes delayed, and is highly significant (P-value: .000). The positive quadratic term suggests that there is a diminishing marginal effect of reducing minutes delayed as share of carrier delay increases. At the critical point of carrier, 53.03 percent, the relationship theoretically switches behavior, which means that an increase in share of carrier delay is associated with an increase in total minutes delayed. However, none of the data points exceed the critical value, which means that this theoretical concern does not occur in reality.

- Holding the sources of delay constant, a 1 percent  increase in the number of scheduled arrivals is associated with a 1.2 percent increase in minutes delayed, and is highly significant (P-value: .003).

As a special note, we also ran several time series models with autoregressive lag terms, and found that these models did not differ from the above regression significantly enough to justify their use. This, along with several other analyses not detailed here, led us to conclude that the heteroskedasticity and auto-correlation, although clearly present and handled via Newey-West errors, do not affect the interpretation or validity of our model.

Discussion

Returning to the analogy proposed under the Model section, these results lead us to conclude that policies aimed at reducing the share of delay due to carriers at O’Hare would be ineffective in reducing overall minutes delayed, and is likely to cause an increase in overall minutes delayed. Further, we propose that a central planner would be most effective in reducing overall minutes delayed by pursuing policies that reduce the share of delay due to other causes, such as late arriving aircraft. These policies could include better traffic management of flights “out-of-order” in the schedule queue, as well as increased co-ordination with airports from which the late flight is departing.

From the perspective of a consumer, these results suggest that any animosity aimed at individual airlines as a cause of delay is unwarranted, and that the most severe delay is a result of factors outside of the consumers’ control, such as weather, or their aircraft arriving late from a prior airport.

These results motivate several areas of further study. The first of which is how various central planning policies affect the sources of delay, and what specific policies should be implemented or discontinued to affect the distribution of delay across delay sources. Secondly, a study which determines the best pre-flight predictors for increases in the different sources of delay would complement our analysis by suggesting a model that does not rely on data acquired only retroactively. Lastly, a comparative study that uses a similar model at an airport with consistently high on-time performance would hold several key results in terms of the differences at high-performance airports (such as SEA, or SLC) to the airport studied herein.

References