Airline safety investments and operating conditions: determinants of aircraft damage severity.

I. Introduction

This study investigates determinants of the severity of damage to aircraft involved in commercial airline accidents and incidents.(1) Airline safety (or the risk of airline travel to the consumer) may be defined as the probability of sustaining injury or death. There are two aspects of this probability: (1) the likelihood of the occurrence of "safety related events" such as airline accidents or incidents (event probability) and (2) the severity of the event given that it has occurred (severity conditional probability). The research heretofore has focused on the former and not the latter aspect of airline safety.(2) However, Moses and Savage |11, 171~ note a limitation of the event probability approach by stating that "even if we were to define safety as the probability that a trip would end in an accident, there still would be the problem that accidents vary in severity from minor damage-only incidents to major tragedies with loss of life." An airline accident, for instance, may involve no deaths or injuries at one extreme to all persons killed and the aircraft destroyed at the other extreme.

This paper contributes to the air safety literature by focusing on the severity of accidents and incidents as an alternative measure of safety and analyzing determinants of severity levels. A micro-data set (from the National Transportation Safety Board) which includes detailed information on individual aircraft accidents and incidents is utilized. Since the severity of an accident is conditioned upon the occurrence of the accident, determinants of the former will be correlated with determinants of the latter. Therefore, analyzing factors that affect severity will yield insight on determinants of the overall level of air safety.

The results of the study suggest that airline safety investments such as more experienced pilots and aircraft safety equipment will significantly reduce event (accident and incident) severity as measured by damage to the aircraft. However, the results indicate that safety investments in newer and larger aircraft and more frequent mechanical and safety inspections will have little effect on aircraft damage severity. The results also indicate that visual flying conditions, the provision of certificated rather than commuter airline services(3) and a reduction in pilot utilization (measured by the number of aircraft landings in a specified time period) are operating conditions that are associated with reductions in the severity of aircraft damage (given that an accident or incident has occurred).

The study is structured as follows. Section II presents a model of aircraft damage severity. Data and estimation results are detailed in section III. The conclusion (section IV) discusses the implications of the findings.

II. A Model of Aircraft Damage Severity

The severity of the aircraft damage of an airline safety related event is a function of airline safety investments and operating conditions. Safety investments consist of actions by an airline to prove the safety of its service; operating conditions describe the environment in which an airline operates.(4) Examples of safety investments by an airline include hiring more experienced pilots, utilizing larger and newer aircraft, utilizing specific aircraft safety equipment and increasing the frequency of aircraft mechanical and safety inspections. Such investments are hypothesized to reduce the severity of a given aircraft accident or incident.

Two measures of pilot experience are included in the analysis: PEXP, the cumulative flight time (in hours) of the first pilot for all commercial aircraft and PAGE, the age of the first pilot in years. More experienced pilots are less likely to make judgmental errors. In addition to lowering the probability of event occurrence, a more experienced pilot is also expected to lessen the severity of an event that does occur. For example, a more experienced pilot may be more adept in recognizing the safety implications of a weather system and the handling of a disabled aircraft. A negative relationship is expected between PEXP and the severity of aircraft damage.

The age of the first pilot (PAGE) is included as an additional measure of pilot experience. The predicted sign of the PAGE coefficient is indeterminant. It is possible that commercial flight time may not include certain types of prior aviation experience. For example, military pilots frequently enter civilian aviation upon retirement at relatively early ages. If so, PEXP will not include this additional aviation experience and hence the relationship between aircraft damage severity and PAGE is expected to be negative. On the other hand, it is possible that advancing age negatively affects pilot performance. This might occur, for example, if age delays the response time of the pilot to in-flight hazards. This would tend to raise the degree of damage for a given event. If, for given levels of PEXP, increasing PAGE reduces pilot performance, the relationship between damage and PAGE is expected to be positive.

The size of an aircraft (AIRWT) is measured by weight in pounds and age of aircraft (AIRAGE) is measured by the cumulative flight time (in hours) on the airframe. As the size of an aircraft (AIRWT) increases, the severity of aircraft damage is expected to decrease, other factors remaining the same. Aircraft size is correlated with airworthiness--larger aircraft are less susceptible to hazardous weather, wind, and other environmental conditions. Therefore, there is a higher probability of involvement by smaller aircraft in a safety related event. Further, a representative event is expected to produce greater relative damage in the lighter aircraft. An accident that may totally destroy a small light-weight aircraft may result in only minor damage to a larger aircraft. AIRAGE is hypothesized to positively affect the severity of aircraft damage. The probability of structural failure increases as the airframe ages, adjusting for other factors.

Variables representing safety characteristics of aircraft are included to test for the impact of safety related equipment and inspection policies on event severity. SSYST is a binary variable indicating the presence of a stall warning system; SSYST equals one if the aircraft has a stall warning system and zero otherwise. To the extent that the functioning of this system provides an early warning of in-flight hazards, a negative effect on event severity is predicted. The variable SINSP is the cumulative flight time (in hours) for an aircraft since the last periodic mechanical and safety inspection. Less frequent inspections raise the probability that a given mechanical or safety problem will deteriorate further and thus increase the probability of greater event severity. A positive relationship is expected between SINSP and aircraft damage severity.

Airline operating conditions that may affect aircraft damage severity include flying conditions, phase of aircraft flight, pilot utilization, and type of airline service. Two measures of flying conditions are considered: VFLY, a binary variable which equals one for visual flying conditions and a zero for instrument flying conditions(5) and DARK, a binary variable which equals one for a night safety related event and zero for a day event. Since poor visibility may impair the pilot's ability to lessen the consequences of a given safety related event, a priori signs for VFLY and DARK are expected to be negative and positive, respectively.

The phase of aircraft flight is represented by the binary variable STRIP, which equals one if the safety related event occurs on an airport's airstrip and a zero for the event occurring off the airstrip. Because the takeoff and landing phases of the flight are relatively more hazardous, a positive relationship is expected between STRIP and damage severity |12~.

Pilot utilization is measured by the total number of aircraft landings by the first pilot during the last 90 days (PUTIL). Takeoffs and landings are recognized as the most stressful portion of flights to pilots. If cumulative fatigue and stress adversely affect pilot performance, then a positive relationship is not only expected between PUTIL and event probability but also between PUTIL and aircraft damage severity.

COMM is a binary variable which equals one if the flight involved commuter service and zero if it involved certificated airline service. This variable is included in the model to control for fundamental differences in the two types of service which may influence event severity. Commuter flights, for example, involve low-altitude, short-hop routes. This implies a greater exposure to hazardous weather operating conditions. Exposure to hazardous weather operating conditions is more likely to result in greater aircraft damage, other factors remaining the same.

III. Data and Results

The data for this study were obtained from the National Transportation Safety Board (NTSB). The NTSB complies detailed data on individual aircraft accidents and incidents that it investigates. NTSB field investigators report the severity of aircraft damage for accidents and incidents as a discrete ranking (DAMAGE). This ranking consists of four classifications: DAMAGE equals one if there was no damage to the aircraft; DAMAGE equals two if there was minor damage to the aircraft; DAMAGE equals three if there was substantial damage to the aircraft; and DAMAGE equals four if the aircraft was totally destroyed. Since our dependent variable (DAMAGE) is an ordinal ranking of aircraft damage severity, we make use of an ordered probit statistical model to investigate determinants of aircraft damage severity.(6) The NTSB data sample includes 339 aircraft accidents and incidents for the years 1983-86 for U.S. scheduled commercial airlines. The sample consists of 200 certificated and 139 commuter airline observations.

Previous researchers addressing determinants of event occurrence have analyzed either certificated service only or certificated and commuter airline service separately. This is based on the view that investigating determinants of pooled certificated and commuter airline accidents, for example, may not be appropriate due to fundamental differences in the nature of the service provided. Further, the use of aggregated data in these studies restrict the availability of explanatory variables to control for differences in the nature of these airline services. By using microdata, this study is able to include such explanatory variables in investigating determinants of the severity of aircraft damage of certificated and commuter airlines.

Perhaps the greatest difference in the aircraft utilized by these airlines is size: commuters tend to be smaller (lighter). The AIRWT variable controls for this differential. With regard to pilot characteristics, commuters generally have younger and less experienced pilots. This is partially explained by the fact that commuter pilot employment is typically an "entry" level position for commercial pilots. By inclusion of certificated and commuter airlines in the severity model, the resulting variation in pilot experience over the career cycle allows for testing of the impact of pilot experience on event severity levels. Finally, safety inspection regulations are less stringent for commuter as opposed to certificated service and certain types of safety equipment (e.g., stall warning systems) are required for certificated but not for commuter airlines. Remaining differences are captured by the binary variable, COMM.

The severity models developed here are designed to test whether variation in safety investment and operating condition variables are associated with the degree of damage to aircraft involved in accidents or incidents. Testing of this hypothesis requires variation in the level of event severity (DAMAGE) and the associated determinants. This necessitates pooling of accident and incident data. Because incidents typically involve minor damage to aircraft (a majority of incidents in our sample were associated with a DAMAGE ranking of 2 with a mean DAMAGE value of 1.67), it is useful to view them as relatively less severe safety related events. Alternatively, accidents may be viewed as relatively more severe events. The mean DAMAGE value for accidents in our sample is 2.85. Further, a small percentage of accidents were associated with minor or no aircraft damage.

Table I. Descriptive Statistics, Mean (Standard Deviation)(*)
Variable            Total sample      Accidents       Incidents
PEXP                 10,511.89         9,255.39       11,635.02
                     (6,224.49)       (6,441.52)     (6,016.79)
PAGE                     39.98            39.06          40.80
                        (12.81)          (10.93)        (14.26)
AIRWT               138,492.91       101,730.74     171,239.31
                   (191,491.44)     (172,775.50)   (201,672.30)
AIRAGE               13,976.64        11,498.22      16,192.00
                    (18,753.00)      (15,602.74)    (20,973.21)
SSYST                     0.94             0.91           0.96
                         (0.25)           (0.29)         (0.19)
SINSP                   172.46           125.74         214.22
                       (745.73)         (687.99)       (793.39)
VFLY                      0.77             0.69           0.84
                         (0.42)           (0.46)         (0.37)
DARK                      0.23             0.27           0.20
                         (0.42)           (0.44)         (0.40)
STRIP                     0.54             0.49           0.59
                         (0.50)           (0.50)         (0.49)
COMM                      0.41             0.53           0.30
                         (0.49)           (0.50)         (0.46)
PUTIL                    38.07            59.43          18.97
                       (100.48)         (123.91)        (68.40)
DAMAGE                    2.16             2.85           1.67
                         (0.93)           (0.95)         (0.48)
N                          339              179            160
* Standard Deviation is in parentheses; N refers to sample
size.

Descriptive statistics (mean and standard deviation) for all variables are presented in Table I. The statistics are presented for the entire sample and separately for accidents and incidents. Note that the percentage of accidents for the two carrier types is roughly equal: 47 percent of all accidents involved certificated carriers and 53 percent involved commuters. For incidents, 70 percent involved certificated carriers versus 30 percent for commuters. The latter statistics may reflect under-reporting of incidents by commuters.(7)

Results of the estimations for several versions of the probit model are presented in Table II. Model 1 includes all variables discussed above; model 2 is identical to model 1 except variables with t ratios of less than 1.0 were dropped (AIRWT, AIRAGE, SINSP, and STRIP). These variables TABULAR DATA OMITTED were deleted because of their lack of significance in the model 1 results. Model 3 consists of the same specification as model 2 but all incidents were deleted from the sample. Model 3 tests stability of the results for accidents only.

Regarding the model 1 results, the signs of all probit coefficients agree with a priori predictions, except for the coefficient sign of AIRWT. The probit equation is significant at the one percent level based upon the likelihood ratio test. Coefficients of safety investment variables PEXP and SSYST are significant at the one percent level, while the coefficient of PAGE is significant at the five percent level. Coefficients of safety investment variables AIRWT, AIRAGE and SINSP are insignificant. The negative signs of the coefficients of PEXP and SSYST suggest that investment by airlines in more experienced pilots and aircraft safety equipment (such as stall warning systems) is expected to reduce the severity of aircraft damage for safety related events. However, the positive sign of the coefficient of PAGE suggests that the use of older pilots (for given pilot experience levels) is expected to increase the damage severity.

The model 1 results indicate that coefficients of the operating condition variables VFLY and COMM are significant at the one percent level, whereas the coefficients of DARK and PUTIL are significant at the five percent level. Results pertaining to variables VFLY and DARK suggest that visual flying conditions are expected to reduce the severity of aircraft damage. Excepting minor differences in the numerical value of the probit coefficients, the model 2 results are basically the same as model 1.

The accidents only model (model 3) does not perform as well as the full sample. The reason that many of the variables become insignificant is that the DAMAGE variable in this model, with a small number of exceptions, takes on values of 3 and 4 only. This model in effect tests whether variation in the explanatory variables is related to the probability of major aircraft damage (DAMAGE = 3) or aircraft destroyed (DAMAGE = 4). Despite this limited variation in the dependent variable, PEXP, VFLY, and COMM coefficients are nonetheless significant at the 5 percent level or better. This demonstrates that the impact of these variables on accident severity is relatively strong and that pooling of accident and incident data, which increase the variation in the DAMAGE variable, is appropriate.

With regard to the main models (1 and 2), the COMM binary variable is highly significant. This finding is consistent with a large volume of accident-rate data which suggest that commuter airline service is less safe than certificated airline service by a wide margin.(8) Note also that the positive sign for the coefficient of the pilot utilization variable (PUTIL) suggests that, adjusting for other factors such as pilot experience, increases in pilot utilization are expected to raise the severity of the aircraft damage for a given event.

The post deregulation period of the U.S. Airline industry has witnessed extensive substitution of commuter for formerly certificated airline routes and a more intensive utilization of pilots (due to service expansions). The results of this study suggest that these developments are expected to have a negative impact on the severity of aircraft damage (and thus air safety) during the post deregulation period.(9)

IV. Conclusion

Findings of this study suggest that safety investments in more experienced pilots and aircraft safety equipment are expected to reduce the severity of aircraft damage. The results produce little or no evidence that investments in newer and larger aircraft and more frequent mechanical and safety inspections will impact safety as measured by event severity. Visual flying conditions, providing certificated rather than commuter airline service and a reduction in pilot utilization are operating conditions that are expected to reduce the severity of aircraft damage.

Opponents of airline deregulation have argued that lower airline profitability expected under deregulated markets will result in a deterioration of airline safety.(10) The validity of this argument depends upon the relationship between deregulation and profits, the extent of the relationships between safety investments and profitability and the extent of the relationships between airline safety and safety investments under deregulation. The findings of this study suggest that certain safety investments are expected to affect airline safety (measured by aircraft damage severity) while others are not. If lower profits in a deregulated environment affect the latter (i.e., negatively affect investing, for example, in newer and larger aircraft and the frequency of mechanical and safety inspections), we would not expect a deterioration of airline safety. If lower profits negatively affect the former (i.e., lower profits negatively affect investing, for example, in more experienced pilots and aircraft safety equipment), we would expect a deterioration of airline safety.

1. The National Transportation Safety Board (NTSB) classifies safety related events that it investigates as accidents or incidents. Accidents are defined 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." An incident is defined as "an occurrence other than an accident, associated with the operation of an aircraft, which affects or could affect the safety of operations." Examples of aircraft incidents include: (1) flight control system malfunction or failure; (2) inability of any required flight crewmember to perform normal flight duties as a result of injury or illness; (3) failure of structural components of a turbine engine excluding compressor and turbine blades and vanes; (4) inflight fire; (5) aircraft collide in flight; (6) damage to property, other than the aircraft, estimated to exceed $25,000; and (7) electrical and hydraulic system failures, loss of power of two or more engines and evacuation of aircraft (where the emergency egress system is utilized) for a large multi-engine aircraft.

2. For example, see Rose |13~, Talley and Bossert |14~ and cited references.

3. A certificated airline is defined by the Federal Aviation Administration (FAA) as "an air carrier operator who conducts operations in accordance with FAR Part 121, who provides scheduled services on specified routes in aircraft with more than 30 passenger seats and a payload capacity of more than 7,500 pounds. These air carriers may also provide non-scheduled or charter service as a secondary operation." A commuter air carrier is defined by the FAA as "an aircraft operator who operates under FAR Part 135 who carries passengers on at least five round trips per week or at least one route between two or more points according to published flight schedules that specify the times, day of the week, and places between which these flights are performed A commuter operates aircraft with 30 or less passenger seats and a payload capacity of 7,500 pounds or less."

4. These terms were introduced by Rose |13~.

5. VFLY may also serve as a proxy for weather conditions.

6. Probit analysis is specifically designed for use in situations in which the true dependent variable is a continuous ordered variable that can not be observed directly but for which ordinal observations (e.g., as those of DAMAGE) representing consecutive sections of the range of the true dependent variable exist. Since our ordinal observations can take on four possible values, we make use of n-chotomous probit analysis, using procedures developed by McKelvy and Zavonia |7~.

7. Since accidents, by definition, are more serious safety related events than incidents, they are difficult to conceal and reporting is likely to be quite accurate. In contrast, the nonreporting of incidents is more difficult to detect. A comparative evaluation of accident and incident data is found in Rose |13, 949~.

8. In 1978, the accident rate (total accidents per 100,000 flight hours) for certificated and commuter airline service was 0.33 and 4.68, respectively; in 1988, these accident rates were 0.29 and 0.93, respectively.

9. It follows by deduction that determinants of the severity of aircraft damage of airline safety related events will also be determinants of those event costs that are positively correlated with the severity of aircraft damage. Research heretofore has investigated the costs of safety related events rather than their determinants. The existence of airline shareholder losses following accidents has been investigated by Borenstein and Zimmerman |1~, Bruning and Kuzma |2~, Chance and Ferris |3~, Cunningham, Slovin, Wood and Zaima |4~ and Mitchell and Maloney |10~. Borenstein and Zimmerman |1~ also estimate airline passenger demand losses.

10. This has been referred to as the profit-safety argument: a reduction in airline profits will result in airlines cutting back on safety investments, which produces a deterioration in safety. The profit-safety hypothesis implies that the market forces will be ineffective in promoting airline safety. The alternative market-response argument suggests that airline safety will not deteriorate under airline deregulation because airlines will anticipate a deterioration in their financial condition following an accident and will take safety precautions in a market environment. The validity of these arguments are discussed in Talley |15~. For a general discussion of U.S. airline deregulation, see General Accounting Office |5~, Kahn |6~, Morrison and Winston |8, 9~ and Moses and Savage |11~.

References

1. Borenstein, Severin and Martin B. Zimmerman, "Market Incentives for Safe Commercial Airline Operation." American Economic Review, December 1988, 913-35.

2. Bruning, Edward R. and Ann T. Kuzma, "Airline Accidents and Stock Return Performance." The Logistics and Transportation Review, June 1989, 157-68.

3. Chance, Don M. and Stephen P. Ferris, "The Effect of Aviation Disasters on the Air Transport Industry: A Financial Market Perspective." Journal of Transport Economics and Policy, May 1987, 151-65.

4. Cunningham, Lawrence F., Myron B. Slovin, Wallace R. Wood and Janis K. Zaima, "Systematic Risk in the Deregulated Airline Industry." Journal of Transport Economics and Policy, September 1988, 345-53.

5. General Accounting Office. Aviation Safety: Measuring How Safely Individual Airlines Operate. Washington, D.C.: U.S. Government Printing Office, 1988.

6. Kahn, Alfred E., "Surprises of Airline Deregulation." American Economic Review, May 1988, 316-22.

7. McKelvy, Richard and William Zavoina, "A Statistical Model for the Analysis of Ordered Level Dependent Variables." Journal of Mathematical Sociology, Summer 1975, 103-20.

8. Morrison, Steven and Clifford Winston, "Air Safety, Deregulation, and Public Policy." The Brookings Review, Winter 1988, 10-15.

9. ----- and -----. The Economic Effects of Airline Deregulation. Washington, D.C.: The Brookings Institution, 1986.

10. Mitchell, Mark L. and Michael T. Maloney, "Crisis in the Cockpit? The Role of Market Forces in Promoting Air Travel Safety." Journal of Law and Economics, October 1989, 329-55.

11. Moses, Leon N. and Ian Savage, "Aviation Deregulation and Safety: Theory and Evidence." Journal of Transport Economics and Policy, May 1990, 171-88.

12. Oster, Clinton V., Jr. and C. Kurt Zorn. "Is It Still Safe to Fly?," in Transportation Safety in an Age of Deregulation, edited by Leon N. Moses and Ian Savage. Oxford: Oxford University Press, 1989, pp. 129-52.

13. Rose, Nancy L., "Profitability and Product Quality: Economic Determinants of Airline Safety Performance." Journal of Political Economy, October 1990, 944-64.

14. Talley, Wayne K. "Deregulation and Safety: The Airline Experience." Unpublished paper, 1991.

15. ----- and Philip A. Bossert, Jr., "Determinants of Aircraft Accidents and Policy Implications for Air Safety." International Journal of Transport Economics, June 1990, 115-30.