Alcohol abuse by airline pilots poses a potential danger to the safety and welfare of the flying public. This paper analyzes two strategies for reducing pilot-error aviation accidents: conducting background checks on pilots for driving-while- intoxicated (DWI) convictions and random preflight alcohol testing of airline pilots. Although both policies have been implemented, no empirical research had previously been conducted to justify either strategy. The results and conclusions of this study are based on analysis of data obtained from the Federal Aviation Administration on the flying performance of 70 164 airline pilots. DWI convictions were found to be associated with a significantly greater risk of a pilot-error accident. In contrast, no evidence was found to validate the assumption that a random alcohol testing program could have prevented accidents. The results provide support for improving the existing DWI background check program and for reducing the sampling rate of random alcohol testing for airline ppilots. This twofold strategy could result in greater improvements in aviation safety and reduced overall costs.
Requests for reprints should be sent to Kathleen L. McFadden, Operations Management and Information Systems Department, Northern Illinois University, McMurry Hall 205, DeKalb, IL 60115-2854.
Two types of alcohol testing are currently in place for airline pilots: postaccident alcohol testing and preflight random alcohol testing. The primary objective of postaccident alcohol testing is to determine whether alcohol was a contributing factor in the aviation accident. Postaccident alcohol testing of pilots began in 1964 and is performed on nearly 100% of pilots involved in airline accidents. However, on September 17, 1991, the U.S. Congress mandated the implementation of random alcohol testing of transportation workers. Random alcohol testing legislation, effective January 1, 1995, was prompted by a series of alcohol-related events in the transportation industry: the 1989 Exxon Valdez oil spill, the 1990 conviction of three Northwest Airlines' pilots for flying under the influence of alcohol, and the 1991 New York subway crash involving an intoxicated engineer. The bill directed the Federal Aviation Administration (FAA) to establish a program that required air carriers to conduct random alcohol testing of airline employees responsible for safety-sensitive functions.
This requirement imposes substantial costs on the airline industry. Initially, employers are required to have 25% of their safety-sensitive workforce randomly tested annually. However, the alcohol testing rule provides a provision for reducing the current 25% sampling rate to 10%, based on evidence that industry violation rates fall below 0.5%. In the aviation industry, the overall theme of the legislation was to prevent potential airline accidents by detecting alcohol use by pilots prior to flight.
A second policy approach--conducting background checks on pilots for driving while intoxicated (DWI) convictions--has been in effect since 1987. Pilots are subject to a $250 000 fine, imprisonment for up to five years, or both for failing to report a DWI conviction to the FAA. In addition, Federal Aviation Regulation 61.15 states that two or more alcohol- and/or drug-related motor vehicle convictions within a three-year period are grounds for the FAA to suspend or revoke a pilot's license. The regulation implies that pilots who have DWI convictions may be unfit to fly, thereby posing a threat to aviation safety. Pilot unions and professional associations have criticized the regulation on the grounds that the FAA had no evidence to support the premise that DWI convictions are in any way linked to flying performance. A concern is that current rulemaking and legislation may be a reaction to public outcry to "do something" rather than a valid means of solving aviation safety issues. According to collective bargaining and arbitration decisions, employers may not interfere in employees' off-the-job behavior unless it adversely affects job performance (Sonnenstuhl, 1989).
This research focused on the determination of whether these two strategies--random preflight alcohol testing and DWI background checks--can be effective in reducing pilot-error aviation accidents. An aviation accident is defined by Title 49, Part 830.2 of the Code of Federal Regulations as "an occurrence associated with the operation of an aircraft which takes place between the time any person boards the aircraft with the intention of flight and all such persons have disembarked, and in which any person suffers death or serious injury, or in which the aircraft receives substantial damage." Specifically, this study focused on the use of these strategies within the airline industry.
Barnett, Abraham, and Schimmel (1979) stressed the need for more empirically based research in aviation safety. Barnett et al. (1979) used the Q statistic to analyze the safety records of 58 major world airlines from 1957 to 1976. Barnett and Higgins (1989) employed the same procedure to update their earlier study. Although the present paper extends the use of statistical analysis in the area of aviation safety, it differs in overall purpose and measurement criteria. The purpose of those prior studies was to scrutinize the safety records of U.S. and international airlines over time. The focus of the current study is not on the safety records of specific airlines but, rather, on the human factor (i.e., on the safety records of the pilots employed by these airlines.)
As part of the study, information from two databases obtained from the FAA were merged: the Civil Aeromedical Institute's Consolidated Airman Information System (CAIS) and the Aviation Standards' Accident/Incident Database (AID). The CAIS database contained information extracted from the National Driver Register (NDR) on pilots who have DWI convictions. On average, convictions remain in the NDR database for seven years. However, information in the NDR was not reliable prior to 1986. Therefore, the scope of this study was limited to the years 1986 to 1992.
Method for DWI Background Checks
A logistic regression model was used to predict pilot-error accidents. The dependent variable was dichotomous (pilot-error accident vs. no pilot-error accident) for the period of the study (1986-1992), and the primary independent variable was XDWI--whether or not the pilot had a DWI conviction or convictions. In the literature, the logistic regression model is identified as a superior method for predicting such binary responses (Hosmer & Lemeshow, 1989). Prior studies in aviation safety (see Li, 1994) have identified the following covariates for possible inclusion in a model that predicts pilot flying performance: (1) XMAJ--whether or not the pilot flies for a major air carrier; (2) XSEX--the pilot's gender; (3) XHRS--the natural log of the pilot's total flying hours; (4) XAGA--the natural log of the pilot's age; and (5) XEXP--the natural log of the pilot's flying hours in the last six months. The following interaction terms were also included in the model: (6) XDWI MAJ--the interaction of XDWI and XMAJ; (7) XDWI SEX--the interaction of XDWI and XSEX; (8) XDWI HRS--the interaction of XDWIand XHRS; (9) XDWI AGE--the interaction of XDWI and XAGE; and (10) XDWI EXP--the interaction of XDWI and XEXP. The log transformations were applied to the continuous independent variables to control for the skewness of the distributions.
Covariate 1 was selected because commuter flights are more likely to crash than are major airline flights (Baker, Lamb, Li, & Dodd, 1993). The U.S. Department of Transportation defines a major airline as an air carrier with operating revenues over $1 billion. In the years 1986 to 1992, these airlines included American, America West, Continental, Delta, Eastern, Federal Express, Northwest, Piedmont, Pan Am Southwest, TWA, United, and USAir. The category of nonmajor airlines includes regional, commuter, and nonscheduled airlines. Prior research also found gender (Vail & Eckman, 1986), total flying hours (Golaszewski, 1983; McFadden, 1993, 1996), age (Golaszewski, 1983; McFadden, 1996), and recent flying hours (Golaszewski, 1983) to be significant variables affecting pilot-error accidents among some pilot groups. No multicollinearity was detected in the sample correlation matrix among the selected variables. The model had the following functional form:
[see article reprint for math]
[see article reprint for math]
None of the interaction terms was found to be statistically significant, nor were the variables XSEX or XEXP. These variables were subsequently dropped from the model. The proportion of pilots with pilot-error accidents was calculated based on the following formula:
[see article reprint for math]
The predicted number of pilot-error accidents for pilots with DWI convictions was [see reprint for math] for the 1372 airline pilots with DWI convictions. The same consideration applied for predicting the number of pilot-error accidents for pilots without DWI convictions.
The logistic regression analysis just presented was designed to provide empirical evidence to either refute or support the relevance of Federal Aviation Regulation 61.15 regarding pilots with DWI convictions. Another important question was whether the FAA should wait until the pilot has received two DWI convictions before intervention. A second logistic regression analysis was performed to address this concern. It was similar to the first model except that the independent variable XDWI was replaced by two variables: XONEDWI (= 1 if the pilot had 1 DWI; = O otherwise) and XTWODWI (= 1 if the pilot had 2 or more DWIs; = O otherwise).
One of the limitations of the database is that it did not include the exact date of the DWI conviction. However, the FAA determined that in 10 of 13 cases in which pilots had both DWI conviction(s) and an aviation accident, the DWI preceded the aviation accident. It has been recommended that the FAA begin including the date of all DWI convictions in their database for subsequent research.
An alpha level of .10 was selected as the acceptable risk for Type I error. This was because of the potential consequences to public safety of failing to conclude a real association between DWI convictions and aviation accidents, and also because of the relative infrequency of both DWI convictions and aviation accidents by professional pilots. Specifically, this level was chosen in order to err conservatively and on the side of passenger safety.
Method for Random Preflight Alcohol Testing
For alcohol testing, the major research question was whether the implementation of a random testing program could have prevented airline accidents by detecting alcohol use prior to flight. Because random preflight alcohol testing was implemented in January 1995, no data were available at the time this manuscript was prepared to allow a direct study of its effectiveness. However, results of postaccident alcohol testing can be used to determine how many accidents might have been prevented had a preflight random alcohol testing program been in effect during the period of study. Postaccident testing will detect whether the pilot had been drinking prior to the accident.
Alcohol testing is performed on all pilots involved in aviation accidents regardless of whether or not the pilot survives the crash. For those pilots who survive, postaccident alcohol testing is performed by local laboratories near the crash site. For pilots killed in aviation accidents, the FAA's Office of Forensic Toxicology and Accident Research Laboratory performs the analysis. They report a near-100% collection rate for pilots. The database was searched to determine the number of airline pilots testing positive for alcohol during postaccident investigations.
Only those pilots classified by the FAA as airline pilots were used in the analyses. Pilots flying for foreign airlines were not included in this study. Of the 70 164 airline pilots studied, 68% (47 758 pilots) flew for a major air carrier and 32% (22 406 pilots) flew for a nonmajor airline. Their average age was 42 years. The majority of airline pilots were male (97.6%) and held a Class 1 Medical Certificate (99.9%); 72% of the pilots held an Airline Transport Pilot (ATP) rating, and the remaining pilots had a commercial rating. The mean total flying hours for these pilots was 8711 h, and the mean recent flight time (last six months) was 311 h. Data include all accidents regardless of whether the pilot was flying in an airliner for employment or in a private plane for off-duty pleasure flying. All pilot performance data were thereby captured.
Results from DWI Background Checks
The first logistic regression model is presented in Table 1. The overall model was found to be useful. DWI was among the significant variables useful in predicting pilot-error accidents.
Figure 1 shows that the model predicted that pilots with DWI(s) would have twice the risk of pilot-error accidents for both major and nonmajor airline categories.
Figure 2 shows the predicted probability of pilot error accidents derived from the second logistic regression model. The model indicated that even one DWI conviction was associated with an increased probability of pilot-error accidents for both major and nonmajor airline pilots. The presence of two or more DWIs almost quadrupled that likelihood.
Number of DWI convictions. Concerning flying performance over the seven-year period studied (1986-1992), 97.55% of airline pilots had neither flying accidents nor DWI convictions, an impressive majority. However, a total of 1372 pilots had DWI convictions, 1.96% of the airline pilot population. The percentages of major and nonmajor airline pilots with DWI convictions are 1.58% (753/47758), and 3.35% (619/18456), respectively. The vast majority (93.4%) of these DWI violators were first-time offenders. Table 2 compares the number of pilot-error accidents for pilots with no DWI, one DWI, and two or more DWI convictions.
Results from Postaccident Alcohol Testing
This study found that no airline pilot tested positive for alcohol during postaccident investigations in the workplace. In other words, from 1986 to 1992, no alcohol-related accidents in airline operations were found. This represents an industry violation rate of 0% on postaccident alcohol testing. However, airline pilots often fly in private planes during off-duty time. Of lesser importance, this study found that three airline pilots were involved in alcohol-related general aviation accidents. General aviation is defined as all civil aviation operations other than airline operations. Basically, it encompasses all private pilots, including off-duty airline pilots who fly for pleasure. These three pilots were employed by nonmajor airlines.
The results from this study showed that DWI background checks could be an effective method for reducing the risk of pilot-error accidents. The presence of even one DWI conviction was associated with a doubling of the risk of a pilot-error accident. These findings indicate that an improved DWI program may be beneficial. The FAA might continue to sanction pilots with two or more DWI convictions but use the first DWI as a trigger to identify and assist the pilot before problems surface in the workplace.
In contrast, this study found no evidence to support the concept of using random preflight alcohol testing as a method for preventing airline accidents. In the seven-year period studied, this research discovered that none of the pilots involved in airline accidents tested positive for alcohol during postaccident investigations. Research conducted prior to 1990 indicates that no U.S. airline accident has ever been associated with alcohol (Modell & Mountz, 1990). Given that no alcohol-related accidents were detected in postaccident testing, it follows that no accidents could have been prevented had a preflight alcohol testing program been in effect during these periods. Therefore, the concept of using random alcohol testing as a method for reducing the number of airline accidents is not supported by this research.
However, the effectiveness of random alcohol testing for its deterrent effect is a separate issue. The data did not reveal the number of airline pilots who might have been drinking prior to flight but were not involved in airline accidents. Random alcohol testing might serve at best as a deterrent for those afraid to drink because it might be detected. Therefore, some degree of random testing may be of value for its deterrent effect. However, the results of this study indicate that a random alcohol testing program is unlikely to reduce the number of airline accidents. Future studies should focus on finding the lowest optimal random alcohol testing rate that does not reduce the deterrent effect or the statistical reliability of the sampling.
DWI background checks are an inexpensive alternative for identifying pilots who have greater risk of involvement in pilot-error accidents. According to the FAA, the cost of verifying DWI information on airline pilots is $175 000 per year, or about $2.50 per pilot. Although the exact cost of random alcohol testing is sketchy at this time, it will be considerably more expensive than DWI background checks. Random alcohol testing involves airline flight delays because the timing of pilot alcohol consumption is of the essence. Taking these delay costs into consideration, the FAA currently estimates that random testing will cost $100 per pilot. Approximately 70 000 pilots will be subject to random alcohol testing. Using cost-benefit analysis, DeLucia and Berg (1993) concluded that the value of random alcohol testing would not justify the costs.
Results of the present study have provided insight and understanding necessary to begin modifying the current approach to dealing with alcohol misuse by airline pilots. The major contributions of this paper have been to demonstrate an association between DWI convictions and flying performance and to show that random alcohol testing is likely to be ineffective in reducing the risk of airline accidents. A reasonable interpretation of the results is that cost-effectiveness and increased airline safety can be realized by improving the DWI program and reducing expenditures on random alcohol testing.
Specifically, the findings provide some justification for lowering the 25% random alcohol sampling rate for airline pilots. However, it may be advantageous to test DWI airline pilots at a higher sampling rate than non-DWI pilots. Moreover, results suggest that the FAA should begin to use DWI information as an intervention technique to detect and assist the potentially risky pilot.
Kathleen L. McFadden is an assistant professor of operations management at Northern Illinois University. In 1993, she received her Ph.D. in business administration with an emphasis in management science from the University of Texas at Arlington.
[Extracted from HUMAN FACTORS: THE JOURNAL OF THE HUMAN FACTORS AND ERGONOMICS SOCIETY, Vol. 39, No. 1., March 1997. Copyright 1997 by the Human Factors and Ergonomics Society, P.O. Box 1369, Santa Monica, CA 90406-1369 USA; 310/394-1811, fax 310/394-2410, email@example.com, http://hfes.