||OR/MS Today, August
When it Comes to Air Travel There's
Safety in Numbers
By Kathleen L. McFadden
OR/MS Models can be used to expand our knowledge of
factors in airline safety and form the basis for setting national and international
Many of us have traveled on an airliner at some point in our lives, possibly to attend an
INFORMS conference or to reach an exciting vacation destination. But studies show that the
majority of passengers aboard flights are not relaxed, but instead are quite afraid of
flying. This may be partially explained by the fact that airline crashes, although rare,
receive a lot of media coverage. So one might ask, "What are my odds of being killed
in an airline accident?" Arnold Barnett, an aviation safety expert at MIT, has found
that the current risk per scheduled U.S. domestic flight is about 1 in 7 million . In
other words, if you picked one flight at random each day, it would take you 19,000 years
before your number was up. These odds sound comforting unless you or your loved one happen
to be one of the unfortunate statistics.
Many individuals in both the public and private sector continually work to improve airline
safety. For example, the operations research department of the Federal Aviation
Administration was established in 1988. More recently, a Center for Excellence in
Operations Research was formed to address various aviation safety issues. While
regulations and strategies are already in place to support the effort of reducing the risk
of aviation accidents and incidents, many more are likely in the future. One major
recommendation of the air safety commission headed by Vice President Al Gore was to strive
to achieve a five-fold reduction in airline accidents over the next 10 years. The role of
academicians in this process is to conduct safety-related research and generate safety
recommendations to help prevent future aviation accidents and incidents. Researchers
routinely analyze accident and incident data and use the information to build practical
tools that are useful in solving aviation safety problems.
The National Transportation Safety Board and the FAA perform investigations subsequent to
an aviation mishap to determine probable cause. Pilot error, maintenance and manufacturing
design flaws are typically cited as cause factors. This discussion focuses on pilot error,
the major cause of aviation accidents. The airline industry is somewhat unique in that the
consequences of even a small error may be fatal. Pilots will inevitably make mistakes. The
challenge for OR/MS professionals is to identify and analyze factors that are associated
with pilot error so that decision-makers might use this information to effectively manage
and reduce that risk.
Factors that influence pilot error accidents and incidents can be broken down into two
major categories as depicted in Figure 1: those that are individual (internal to the
pilot) and those that are situational (external to the pilot).
In the past most researchers have relied on only rudimentary statistical methods in
analyzing accident and incident data. Recently, researchers have begun to see the
potential of OR/MS techniques in contributing to the solutions of aviation safety
problems. Advanced modeling offers a richer, more comprehensive approach to addressing
aviation safety concerns. The new thrust is to expand the range of models and statistical
tools that are used to analyze safety data. My research has applied OR/MS modeling
techniques and has focused on the individual factors associated with pilot error accidents
and incidents, specifically gender and medical (alcohol misuse). Policy-relevant aviation
research coupled with the use of advanced modeling, as opposed to rudimentary statistical
methodologies, can provide the information needed to have an impact on the decision-making
process. My intent is to illustrate this point using examples.
General and Pilot Error Accidents
Females have recently become a more significant group in the airline pilot profession. It
has been argued that differences (e.g. physical, physiological, psychological) between
males and females may affect their flying performance. It is well established in the
literature that a wide variety of aptitudes, skills and cognitive abilities differ among
the sexes. The largest cognitive gender differences are found in visual-spatial abilities.
Research shows that males have greater visual-spatial skills than females. Males also tend
to be superior in the quantitative area, while females tend to have better verbal skills.
Cognitive performance and spatial abilities are among the most important attributes of
flying. Verbal skills are also important to maintain safe air traffic control
communication and facilitate crew coordination.
To address possible gender differences in pilot flying performance, I first used
contingency table analysis. With this simple approach, I found that female airline pilots
were significantly more likely to have aviation
accidents than their male counterparts. After exploring the data further, I discovered
that female airline pilots, on average, were less experienced and much younger than males.
Studies of the effect of age on pilot error accidents have demonstrated that accident
rates decrease with age, but may level off for older pilots. Accident rates also tend to
decrease as experience (measured by total flying hours) increases. Therefore, it was
important to use a more sophisticated modeling technique that could address the issue of
confounding of factor effects. Since males were older and more experienced, this explained
their lower accident rates.
The differences in age and experience levels of males and females were due to the fact
that females have only recently entered the airline pilot profession in any significant
number. After adjusting for variables in a logistic regression model, accident rates of
male and female airline pilots were not significantly different . These findings
suggest that airlines should recruit and retain experienced pilots regardless of gender.
It also cautions against affirmative action programs that lower the flying standard for
females in order to increase the number of female airline pilots. More research could be
done in this area using OR/MS models. For instance, it may be worthwhile to analyze other
measures of flying performance for gender differences. Incidents, pilot deviations and
simulator check-ride performance errors occur with greater frequency than accidents and
may provide higher statistical power.
Future studies might also compare the flying performance of those pilots trained in the
military versus those trained through the civilian ranks.
As OR/MS professionals, it may seem somewhat obvious that important factors should be
adjusted for in a model. However, some prior aviation safety researchers have based their
conclusions on simple statistical tests that may provide misleading results. As an
example, one published gender study in aviation  was criticized because it used a
simple approach that failed to control for recent flying hours, a measure of exposure of
pilots to risk. The theory behind adjusting for risk exposure is that pilots who fly more
frequently may be exposed to a greater risk of being involved in an accident. The study
reported that females flying in general aviation (private flying) had significantly lower
accident rates than males, and were a safer pilot group. But, if males had more recent
flight time, this could explain their higher accident rates. More advanced modeling
techniques should be applied to the data in general aviation to confirm prior findings of
Alcohol Misuse and Pilot Error Accidents
Another one of my projects involved analyzing two strategies for reducing pilot-error
aviation accidents: conducting background checks on pilots for driving-while-intoxicated
(DWI) convictions and random alcohol testing of airline pilots . DWI background checks
have been in effect since 1987, while random alcohol testing began in 1995. Although both
policies had been implemented, no empirical research had previously been conducted to
justify either strategy. Random alcohol testing has imposed substantial costs on the
airline industry. In contrast, the cost of verifying DWI information has been quite
inexpensive, only about $2.50 per pilot.
The FAA may suspend or revoke a pilot's certificate or rating if the pilot has two or more
DWI convictions within a three-year period. The FAA verifies DWI information on pilots by
querying the National Driver Register. Information in the NDR was unreliable prior to
1986. Therefore, the scope of my study was limited to the years 1986-1992.
I found that the vast majority of airline pilots (97.55 percent) had neither flying
accidents nor DWI convictions over the seven-year period. However, 1,372 pilots had DWI
convictions, 1.96 percent of the airline pilot population. Table 1 compares the number of
pilot-error accidents for pilots with no DWI, one DWI and two or more DWI convictions.
|To address the association between DWI
convictions and pilot flying performance, I applied loglinear modeling techniques. The
primary advantage of this approach was that it allowed me to access multiple factors
simultaneously (age, experience, gender, risk exposure and major/non-major airline
employment) and to partition the contribution to variance of DWI over and above the
contribution of the other factors. The results showed that DWI was significant even after
adjusting for important factors. The presence of even one DWI conviction was associated
with a doubling of the risk of pilot-error accidents. The presence of two or more DWIs
almost quadrupled that likelihood. In contrast, my study found no evidence to support the
concept of using random preflight alcohol testing as a method for preventing airline
accidents. The findings suggest that cost-effectiveness and increased airline safety could
be realized by improving the DWI program and reducing expenditures on random alcohol
I presented recommendations for policy improvements to both the FAA and the NTSB based on
these findings. They were also made part of the public docket on the alcohol testing rule.
My first recommendation was that the FAA continue to penalize pilots with two or more DWI
convictions but use the first DWI as a trigger to identify and assist the potentially
risky pilot. My second recommendation was to reduce the current random alcohol testing
sampling rate based on these findings. This twofold strategy could result in greater
improvements in aviation safety and reduced overall costs.
Future Research and Directions
It has been estimated that airline travel will increase at a rate of about 5 percent per
year. While our nation's current level of air safety is remarkably high, simply
maintaining that level may be inadequate for the future. Aviation safety research is
critical to our success in coping with this rapidly expanding environment. Examples of
just a few current and proposed research activities include the following:
|Developing an information infrastructure, known as the Global Analysis and Information
Network (GAIN), for collecting, analyzing and disseminating aviation safety information.
|Developing technologies to be used in cockpits and ground control systems that reduce
pilot error and pilot complacency. |
|Examining trends in national and international airline safety data using new
|Developing a computer program to study the errors of airline maintenance workers and
determine solutions. |
|Examining the challenges and opportunities surrounding global airline mergers and
Barry Valentine, the former acting FAA administrator, has stated that the FAA, "is
committed to continually working to make the safest air transportation system in the world
even more safe." Researchers are also committed to keeping one step ahead of change
in this dynamic world of aviation, and OR/MS models may well be the key to achieving this
I hope to have conveyed how valuable OR/MS skills are in the area of aviation safety.
Aviation safety research offers exciting opportunities for OR/MS professionals. The
aviation industry and the government are moving toward a more heightened technological
environment. As they begin to incorporate more rapid and advanced database systems that
collect and store a vast amount of information, it becomes essential to effectively
analyze the data to help prevent future mishaps. This is the area where people trained in
OR/MS can make valuable contributions.
OR/MS models can be used to expand our knowledge of factors in airline safety and form the
basis for setting new, important national and international policies. As we become more
involved in aviation safety issues, a broader range of alternative and more sophisticated
modeling techniques can be applied to aviation safety research which in turn will improve
the safety of our nation's air transportation system.
1. Barnett, A., "Aviation Safety, Another Decade,"
Presentation at Informs San Diego, May, 1997.
2. McFadden, K. L., "Comparing Pilot-error Accident Rates of Male and Female Airline
Pilots," Omega, Vol. 24, No. 4, 1996, pp. 443-450.
3. Vail, G. J. and Eckman, L.G, "Pilot-error Accidents: Male vs. Female," Applied
Ergonomics, Vol. 17, 1986, pp. 297-303.
4. McFadden, K.L., "Policy Improvements for Prevention of Alcohol Misuse by Airline
Pilots," Human Factors, Vol. 39, No. 1, 1997, pp. 1-8.
Kathleen L. McFadden, Ph.D., is an assistant professor of Operations
Management at Northern Illinois University, DeKalb, Ill.