You are on page 1of 12

HUMAN FACTORS, 1988,30(3),293-304

Multiattribute Modeling Analysis of the


Effects of a Low Blood Alcohol Level on
Pilot Performance

LEONARD E. ROSS' and JAMES C. MUNDT, University of Wisconsin, Madison, Wisconsin

Multiattribute modeling procedures were used to evaluate the flight performance of pilots
who completed a simulator flight under 0 and 0.04% blood alcohol concentration (BAC)
conditions. The flight involved VOR tracking, vectoring, traffic avoidance, and descent.
Flight instructors' judgments were used to develop a multiattribute model of flight perfor-
mance that permitted evaluation of the effects of alcohol on overall flight performance, as
well as on task segment and performance aspect components of the flight. Alcohol was
found to have a significant deleterious effect on overall pilot performance and on some of the
task segments. The multiattribute modeling approach was found to be useful in providing a
task analysis function that permitted alcohol effects to be evaluated in a manner that re-
flected pilot concentration on some aspects of the flight task at the expense of others.

INTRODUCTION equipment to measure any ingested alcohol"


(1984, p. 21). Interest in the effects of lower
It is commonly recognized that flying with BAC values on pilot performance has been
a high blood alcohol concentration (BAC) is heightened by the FAA's 1985 adoption of a
very dangerous and often fatal. This conclu- rule that no person with a BAC of 0.04% or
sion is strongly supported by accident data higher may act or attempt to act as a crew
and by research that has demonstrated a va- member of a civil aircraft.
riety of performance decrements as BAC ap- The experimental evidence with respect to
proaches and exceeds 0.10%. With respect to the effects of low BAC on pilot performance is
lower BAC values, common sense suggests not extensive, however, and in some cases is
that the presence of any alcohol in a pilot's somewhat contradictory. In proposing and
blood can jeopardize flight safety. The Na- adopting the 0.04% BAC value the FAA cited
tional Transportation Safety Board has taken only one study, that by Billings, Wick, Gerke,
that position and recommended in 1984 that and Chase (1972). In this study eight experi-
the Federal Aviation Administration (FAA) enced and eight less experienced pilots flew
set the BAC limit "at the lowest possible level ILS approaches at night in a Cessna 172
consistent with the capability of testing while under the influence of each of four
blood alcohol levels: 0.0%,0.04%,0.08%, and
I Requests for reprints should be sent to Leonard E.
0.12%. Significant alcohol effects were ob-
Ross, Department of Psychology, University of Wisconsin,
1202 W. Johnson St., Madison, WI 53706. served with the higher BAC values for lateral

© 1988, The Human Factors Society, Inc. All rights reserved.


294-June 1988 HUMAN FACTORS

ILS control in the case of experienced pilots, Mundt (1986) found that pilots with a 0.04%
whereas less experienced pilots showed sig- BAC showed decrements in some simulator
nificant decrements in both lateral and verti- flight measures while maintaining straight
cal ILS control. Minor decrements in ILS and level flight under light and moderate
tracking were observed at the 0.04% BAC turbulence conditions. In this study pilots
value only for the less experienced pilot monitored front and side displays for aircraft
group. Most important, both groups were ob- silhouettes representing aircraft targets dur-
served to make linearly increasing numbers ing flight and were found to make fewer
of major procedural errors at BACs of 0.04%, scanning head movements when under the
0.08%, and 0.12%. influence of a 0.04% BAC. In the final stage of
The results of simulator studies concerned the flight pilots were given an additional task
with the effects of low BAC values on pilot of either copying weather information for
performance have been mixed. Aksnes (1954) their destination airport or making commu-
reported that BACs of about 0.05% had dele- nication and navigation radio frequency
terious effects on pilots' abilities to fly a Link changes, during which the simulator's in-
trainer over a U-shaped course requiring alti- struments and a horizon displayed on a front
tude and airspeed changes, and suggested monitor showed a banked climb or descent
that even lower BACs could have an effect- that gradually increased over a 5-s period.
that is, "one must be prepared for a lower Under the 0.04% alcohol condition the sub-
ability to perform with a 0.02 percent B.A." jects lost, in the case of a banked descent, or
(p. 688). Henry et al. (1974) also concluded gained, in the case of a banked climb, signifi-
that BACs of 0.02-0.035% could produce sig- cantly more altitude than without alcohol
nificant decrements in performance of a sim- and also took longer to recover from their un-
ulated cross-country flight involving 37 usual attitudes. Two pilots with a 0.04% BAC
maneuvers over a one-hour test session. in effect "crashed" the simulator during this
Interpretation of these data is equivocal, flight segment.
however, given that only 2 of the 22 subjects Overall, these results suggest that the ef-
were licensed pilots, not all measures showed fects of a low BAC may become apparent only
significant differences at this BAC range, and under certain test conditions. It appears that
the experimental procedures make data anal- they are most likely to be found when multi-
ysis difficult. ple tasks are required of the subject, or per-
More recently Taylor, Dellinger, Schilling, haps when the primary task load is very
and Richardson (1983) tested eight pilots in a heavy. In a relatively focused task such as
simulator flight task that involved holding keeping glideslope and localizer needles cen-
patterns and ILS approaches. Performance tered, the effects of low blood alcohol levels
decrements were significant for subjects may not be apparent, but even a moderate or
tested with a median BAC of 0.082% but not low BAC may lead to performance decre-
with BAC values of 0.038% or 0.014%. Using ments when multiple tasks with distributed
the same task, Hyman, Taylor, Weller, and attentional demands must be performed. In
Nagel (1986) failed to find significant overall some cases attention may be inappropriately
effects of a 0.05% BAC on pilot performance. distributed among tasks in a manner that
In both of these studies the subjects were could be quite dangerous in the flight envi-
given a secondary (Sternberg) task to per- ronment-for example, when primary atten-
form while "flying" the simulator. tion is allocated to a task that is secondary in
In contrast to these results, Ross and importance to flight safety.
ALCOHOL AND PILOT PERFORMANCE June 1988-295

If measurement of more complex combina- screened for any physical condition or drug
tions of tasks such as those often found in ac- use that contraindicated alcohol consump-
tual flight is required to accurately assess tion. Those not using alcohol or attempting
alcohol's effects, then it may be necessary to to abstain from alcohol use were not ac-
devise and use multiattribute measures of cepted as subjects.
pilot performance. In the present study,
Apparatus
pilots' performance on a cross-country simu-
lator flight was compared under 0.0% and A highly modified ATC 610 simulator (pro-
0.04% BAC conditions. A multiattribute per- cedures trainer) was used in the study. The
formance model, based on multiattribute simulator was in a specially constructed en-
value and utility assessment techniques dis- closure that contained a front and two side
cussed by Keeney and Raiffa (1976) and von monitors with screen sizes of 42 cm x 31 cm
Winterfeldt and Edwards (1986), was devel- and 27 cm x 21 cm, respectively. The side
oped to assess performance. In this model, monitors were positioned such that a hori-
single dimensional flight parameters were zontal head movement greater than 20 deg
weighted and combined into multidimen- was required to detect targets displayed on
sional aspects of pilot performance. These the screen. A horizon (with a filled "ground"
performance aspects were, in turn, weighted area) was always present on the front screen,
and combined into task segment scores, with the horizon tilting and moving verti-
which were then combined to provide an cally in synchrony with the horizon line of
overall index of pilot performance. Use of the the attitude indicator. Small silhouette air-
evaluation model permitted examination of craft targets could appear at any of 8 loca-
pilot performance in terms of these perfor- tions on the top part of the front monitor
mance aspect and task segment scores, and screen or in any of 16 locations on either of
also provided a procedure by which an over- the side monitor screens. Once they ap-
all index of flight performance could be de- peared, the silhouettes were present for 40 s
rived. and either remained constant in size or be-
came larger (loomed) to indicate an ap-
METHOD proaching aircraft. The subject pressed a
button located on the left horn of the control
Subjects
yoke to indicate detection of the target.
Twelve male pilots (median age = 27 The simulator was modified such that the
years, range = 19-39 years) served as sub- instruments were backlighted and many of
jects in the experiment. All had a relatively the electronic components were replaced
low number of flying hours (median = 110, with higher-precision items. The aircraft seat
range = 50-315), and none had an instru- used with the simulator could be adjusted
ment rating or a significant amount of in- vertically and the rudder pedals could be ad-
strument training. Six had served in a pre- justed to the subject's preferred position.
vious study using the same simulator but a Each subject's head position was the same
different flight task, which they had flown distance from the instrument panel and
with both a BAC of 0.0% and a BAC value monitors. Head movements were measured
similar to that used in this study. Thus these using a helmet device that permitted move-
subjects could be considered somewhat expe- ment in the horizontal and vertical planes.
rienced with respect to both simulator flight Transducers attached to the helmet provided
and the alcohol variable. Subjects were head orientation information to O.1-deg accu-
296-June 1988 HUMAN FACTORS

racy. The BAC determinations were made changes were also required, and "air traffic"
from breath samples by a Federal Signal Cor- in the vicinity of the pilot was reported. On 9
poration Model 5000 Intoxilizer. of the 13 occasions when a traffic call was
Subjects' responses and flight performance made, an aircraft silhouette target was pre-
measures were sampled, digitized, recorded, sented on one of the three monitors, and the
and analyzed by computer. The measures pilot was instructed to press the control yoke
used and their sampling intervals were as button as soon as the target was sighted. Ap-
follows: target detection button press, hori- pearance of the targets was delayed up to 1
zontal head movements, and pitch and roll min after the call. If the target loomed to-
yoke control movements, 100 ms; attitude in- ward the pilot, which occurred four times
dicator pitch and roll, vertical speed indica- each flight, he was instructed to make a colli-
tor (VSI), and course deviation indicator sion avoidance maneuver; that is, a descent
(CDI), 200 ms; and altimeter, directional gyro of 200 ft (60.96 m) if the looming target ap-
readings, and airspeed, 500 ms. peared in the upper half of the monitor
screen, or a climb of 200 ft if the target ap-
Design
peared in the bottom half of the screen.
Each subject was tested under 0.0% and Looming and non looming targets appeared
low (0.04%) BAC conditions on two consecu- on the side screens, but only nonlooming tar-
tive days according to a Latin square design. gets were presented on the front screen. The
Two flight tasks were used that were similar subject was not informed of that fact prior to
in that the same flight tasks were involved the flights.
but with different headings, altitudes, and
Procedures
sequence of traffic targets. Half of the sub-
jects received one order of the two flights, Following completion of screening ques-
while the other half received the reverse tionnaires, subjects were given either alcohol
order across the two experimental flight and orange juice or a placebo condition in
days. which 3 cc of alcohol was floated on top of
Subjects were given 90 min of instruction orange juice. Three 150-cc drinks were ad-
and simulator practice on the day prior to ministered over a 25-min period that started
the first day of the two-day experimental se- 50 min prior to the beginning of the test pe-
quence. In addition, subjects received a 15- riod, with subjects given 5 min to complete
min practice flight prior to the beginning of each drink. A 96% ethanol dose of 0.5 gm/kg
each experimental flight. The simulator of body weight was found necessary to obtain
flight task required the subject to complete a a 0.04% BAC at the time of testing. A very
preflight checklist, intercept and track a VOR small (0.062 gmlkg) maintenance drink of al-
radial inbound to a VOR, turn to a new radial cohol in orange juice (or a placebo drink for
and follow it outbound, and then continue nonalcohol days) was also given to subjects
the flight until instructions were received to just before they entered the simulator to en-
descend preparatory to landing. A landing sure that their BAC would remain close to the
checklist was then to be used, followed by de- desired level over the experimental period
scent to an assigned altitude. Recorded con- (Lentz and Rundell, 1976). The subject's BAC
troller messages were given throughout the was determined when he entered the labora-
45-min flight, and at several points the con- tory, prior to entering the simulator, at the
troller vectored the pilot away from and back end of the flight period, and regularly there-
to the radial. A number of radio frequency after until BAC reached 0.0%, at which time
ALCOHOL AND PILOT PERFORMANCE June 1988-297

the subject was released. Subjects in the pla- resulted in the identification of three perfor-
cebo group on Day 1 were kept for several mance aspects: Vertical Control, Horizontal
hours after the flight and given several Control. and Cockpit Duties. These aspects
breath tests during that time so they would were considered by the instructors to be or-
not become aware that they had not received thogonal dimensions of pilot performance,
alcohol. Although every attempt was made to and the flight parameters that indexed these
keep the subjects unaware of the amount of dimensions were judged to be value-indepen-
alcohol they consumed, all were aware from dent in terms of their contributions to the
internal cues and self-perceived effects of al- identified performance aspects, although it
cohol on their behavior that they received was recognized that they are not orthogonal
more alcohol on the alcohol day than on the in terms of flight characteristics (e.g., roll
placebo day. changes will affect aircraft heading and
therefore course deviation indications).
The Multiattribute Performance Model
Several days later the instructors were
Multiattribute value and utility assess- provided with a detailed time line and tran-
ment techniques (Keeney and Raiffa, 1976; scripts of the flight communication sce-
von Winterfeldt and Edwards, 1986) were narios, and were asked to identify flight seg-
used to develop the model structure and to ments in which fundamentally different tasks
quantify the elements within the model. The were required of the pilots. These task seg-
two experienced flight instructors whose ments were to be unique in terms of the
judgments were used in developing the weightings of their component performance
model were informed that the purpose of the aspects, or the selection of the flight parame-
model was to create a scale of pilot perfor- ters from which the performance aspects
mance that could be used to score flight per- were computed.
formance as they would have had they actu- The flight measures were scaled in the fol-
ally observed the flight. The range of lowing manner. For each flight parameter
performance of interest was described as on a within each performance aspect of each task
continuum between a theoretically perfect segment, the instructors were asked to pro-
flight and a flight that would have a signifi- vide an optimal value and the deviations
cant probability of an accident. The instruc- above or below that value that, if exceeded,
tors were familiarized with the equipment would lead to a situation that had a signifi-
and procedures and allowed to fly the same cant probability of an accident occurring.
flight scenario as the subjects. They were then asked to draw a continuous
The first step was the compilation of a list curve on a graph such that the optimal value
of the flight parameters that would be useful received a value of 100 and the maximum de-
in evaluating pilot performance-that is, al- viation received a value of O. Using these
titude, heading, reaction time to targets, and functions the sampled data, averaged over
so on. After the instructors were satisfied that 5-s intervals, were converted into flight pa-
the list was exhaustive, they were asked if the rameter values between 0 and 100, inclusive.
various flight parameters could be naturally The importance of the different flight pa-
grouped with respect to the type of perfor- rameters contributing to the performance
mance information they represented. They aspects was determined using a ratio estima-
were also asked to identify any flight parame- tion technique. For each task segment the in-
ters that were redundant with respect to the structors were asked to identify the most im-
information they conveyed. This procedure portant flight parameter within each of the
298-June 1988 HUMAN FACTORS

performance aspects, and this flight parame- nition, the expected utility of such reference
ter was assigned a value of 100. The remain- lotteries is 0.5, the specific value at which the
ing flight parameters within the performance instructors expressed indifference was also
aspects were assigned numbers reflecting assigned a utility of 0.5 and a utility function
their importance relative to the parameter was approximated.
that had been assigned the value of 100. The The second question concerned the possi-
numbers assigned to the flight parameters bility that components at a given level of the
within each performance aspect were then model might interact in indexing pilot per-
summed, and the number assigned to each formance, thus requiring multiplicative
individual parameter was divided by this rather than additive functions. Because it
sum. This produced scale constants that re- was not known how, or if, such combina-
flected the relative importance of the flight tional differences would affect the validity of
parameters constituting each performance the model, evaluation of pilot performance
aspect. For each 5-s interval, the derived was conducted using both combinational
flight parameter values were multiplied by forms for both value and utility measures.
their scale constants (which depended on the Multiplicative functions were derived
flight task during that interval) and then through the use of procedures described by
summed, yielding an aspect value for each of Keeney and Raiffa (1976). Additive functions
the three performance aspects for each 5 s of were derived using the ratio estimation tech-
the flight. If any flight parameter within a nique described earlier for combining flight
performance aspect had a value of 0, then the parameter values into performance aspect
performance aspect value for that interval values. These quantitative forms of the model
was also made O. were computed at both the performance
Two questions were considered concerning aspect to task segment score and the average
the combining of performance aspect values task segment to overall pilot performance
into task segment scores and task segment measure levels of the model (see Mundt and
scores into overall pilot performance mea- Ross, 1987, for additional discussion).
sures. The first concerned whether the values
obtained from the derived functions should RESULTS
be converted into utility measures and thus
Blood Alcohol Levels
incorporate the instructors' attitudes toward
risk, or whether these attitudes were embed-
The mean BAC for pilots entering the simu-
ded in the value judgments made by the in-
lator was 0.043% with a standard deviation
structors during construction of the model.
(SD) of 0.004% and a range of 0.036%-
Consequently, quantitative forms of the
0.048%. At the end of the flight the mean BAC
model using both values and utilities were
was 0.038% with an SD of 0.003% and a range
constructed.
of 0.033%-0.045%.
Utility functions were constructed by de-
termining a specific performance aspect The Multiattribute Performance Model
value, or a task segment score, to which the
instructors were indifferent in comparison The four different task segments that were
with a lottery that had a 0.5 probability of identified were VOR Tracking, Vectoring, De-
that aspect or task segment being at either scent, and Collision Avoidance, each of which
maximum or minimum. Given that, by defi- was indexed in terms of the three perfor-
ALCOHOL AND PILOT PERFORMANCE June 1988-299

mance aspects described earlier. The overall to the CDI (course setting), checklist, scan-
model structure is shown in Figure 1. ning, and target detection. If the subject had
The three performance aspects identified no trouble with the OBS course setting
were horizontal control of the aircraft, verti- throughout the flight, a score of 100 was as-
cal control of the aircraft, and cockpit duties signed. If there was apparent trouble in set-
required of the pilot. Horizontal Control was ting the OBS but the correct setting was
defined for each 5-s interval of each task seg- achieved, a 50 was assigned. If experimenter
ment in terms of the following performance intervention was required to obtain the cor-
measures: heading, roll, CDI deviation, and rect setting, the subject received a O. If the
aileron control input. Heading was scored as checklist was used without difficulty, a 100
the average deviation between the actual and was given, but if a problem occurred, a a was
expected heading, roll as the average bank received. Scanning was scored by assigning
angle, CDI deviation as the average deviation the subject a 100 if a head turn occurred in a
from the radial in degrees, and aileron con- 5-s interval, and 4 points less than the value
trol as the SD of the yoke-roll A to D samples. for the previous interval if a head turn did
Vertical Control was defined for each 5-s in- not occur. This scoring scheme was consis-
terval in terms of altitude, pitch, airspeed, tent with the instructors' judgments that
vertical speed, and elevator control input. Al- scanning for other aircraft at least every 5 s
titude was scored as the average deviation was optimal and that neglecting to scan for
from the correct altitude, pitch as the aver- more than 2.5 min should result in a value of
age pitch angle, airspeed as the average de- O. Finally, for target detection, the subject re-
viation from the assigned airspeed of 110 ceived 100 minus 2.5 times the number of
knots (90 knots on descent), vertical speed as seconds the target was presented but not de-
the average VSI value, and elevator control tected. Thus targets detected immediately re-
input as the largest absolute change between sulted in a value of lOa, and missed targets
two A and D samples within the 5-s interval. produced a value of 0 because targets were
Flight parameter deviations from optimal present for 40 s.
levels for both the alcohol and no-alcohol The output of the different quantitative
conditions are shown in Table 1. forms of the model (value versus utility mea-
Cockpit Duties were defined with respect sures combined additive1y or multiplica-
tively) was compared for differences.
PILOT PERFORMANCE Whereas generally the same alcohol effects
5 were significant with utility functions, their
"I
----I--t--I----1 E
T G added complexity, combined with the ques-
VOR
TRtlNG f
VECTORING
f
DESCENT COLLISION
AVOiANCE
A
5 E
K N
T
5
M
tions raised by von Winterfeldt and Edwards
(1986) concerning the distinction between
p
VERTICAL
CONTROL
VERTICAL VERTICAL VERTICAL E value and utility measures, suggested that
CONTROL CONTROL CONTROL R A
F 5 value models would be appropriate for mea-
HORIZONTAL HORIZONTAL HORIZONTAL HORIZONTAL 0 P
CONTROL CONTROL CONTROL CONTROL R E
COCKPIT COCKPIT
M C
A T
suring pilot performance in the present
COCKPIT COCKPIT
N S
DUTIES DUTIES DUTIES DUTIES
C study. Therefore, only the results from the
E
t t
FLIGHT PARAMETERS
t t value-additive and value-multiplicative
models will be presented.
Figure 1. Pilot perfonnance evaluation model struc- The additive model weightings of the task
ture. segment value for the overall pilot perfor-
300-June 1988 HUMAN FACTORS

TABLE 1

Flight Parameter Deviations from Optimal Levels for Vertical and Horizontal Control Performance
Aspects Averaged over Task Segments

Collision
VOR Vectoring Descent Avoidance

Alc No Alc' Ale No Ale Ale No Ale Ale No Alc

Vertical Control
Altitude 82 75 47 47 127 114
Pitch 1.5 1.4 1.4 1.4 2.3 1.9 2.6 2.4
VSI 250 228 212 205 486 413 431 436
Airspeed 4.3 3.5 4.7 4.2 9.2 8.8
Elevators 22.5 17.6 23.7 18.3 20.1 16.7 22.0 17.5
Horizontal Control
Heading 13.9 11.1 20.6 15.2 14.4 10.4 19.6 12.7
Roll 5.5 4.9 6.6 6.3 5.9 5.4 7.0 5.9
COl 2.0 1.4 2.7 1.2 4.1 2.7
Ailerons 51.7 64.1 50.5 37.7 47.9 43.3 45.0 43.2

• Data from one subject excluded because of descent below 0 fl.

mance score were as follows (the abbrevia- task segment measures, the weightings of the
tion for each term is given following the performance aspect values for the additive
term): model were:

Performance = 0.243 v(VOR) Performance = 0.5 v(Vertical Control, VC)


+ 0.243 v(Vectoring, Vec) + 0.375 v(Horizontal Control,
+ 0.243 v(Descent, D) HC)
+ 0.270 v(Collision Avoid- + 0.125 v(Cockpit Duties, CD)
ance, CA)
and for the Collision Avoidance task segment:
For the multiplicative model the corre-
sponding values were: Performance = 0.364 v(VC) + 0.364 v(HC)
+ 0.272 v(CD)
Performance = 0.85 v(CA) + 0.255 v(D)
+ 0.425 v(Vec) + 0.425 v(VOR) For the multiplicative model the VOR and
- 0.205 v(CA)v(D) Vectoring weights were:
- 0.342 v(CA)v(Vec)
Performance = 0.825 v(VC) + 0.124 v(HC)
- 0.342 v(CA)v(VOR)
+ 0.008 v(CD)
- 0.103 v(Vec)v(D)
+ 0.04 v(VC)v(HC)
- 0.103 v(VOR)v(D)
- 0.171 v(VOR)v(Vec)
For Descent:
+ 0.083 v(CA)v(D)(Vec)
+ 0.083 v(CA)v(D)v(VOR) Performance = 0.70 v(VC) + 0.05 v(HC)
+ 0.041 v(D)v(Vec)v(VOR) + 0.007 v(CD)
+ 0.138 v(CA)v(Vec)v(VOR) + 0.01 v(VC)v(CD)
- 0.033 v(CA)v(D)v(Vec)v(VOR) + 0.172 v(VC)v(HC)
In the case of VOR, Vectoring, and Descent For Collision Avoidance:
ALCOHOL AND PILOT PERFORMANCE June 1988-301

Performance = 0.95 v(VC) + om v(HC) and the order of alcohol administration (alco-
+ 0.475 v(CD) hol versus no alcohol on Day 1) as between-
- 0.462 v(VC)v(CD) subjects factors. The overall pilot perfor-
mance scores resulting from the additive
Average additive model scores for overall
model showed a significant alcohol perfor-
pilot performance, task segments, and perfor-
mance decrement, F(1,8) = 5.32, P < 0.05. A
mance aspects for both the alcohol and no-al-
significant interaction of alcohol and alcohol
cohol conditions are shown in Table 2.
order, F(1,8) = 7.90, P < 0.025, and a signifi-
cant interaction of alcohol, alcohol order,
Analysis of Performance Measures and flight sequence, F(1,8) = 8.84, P < 0.02,
were also found. The Alcohol x Alcohol
The overall performance scores of the sub-
Order interaction reflected the fact that the
jects who had participated in a previous sim-
alcohol decrement was greater when alcohol
ulator study were compared with the scores
was received on Day 1 than when it was re-
of those who had not. No significant differ-
ceived on Day 2. The interaction of alcohol,
ences were found involving this factor, so it
alcohol order, and flight sequence appeared
was not considered in the following analyses.
to be a result of the greater difficulty of the
An analysis of variance was carried out on
Flight B sequence, which, when flown on Day
the model-derived performance scores with
1 without alcohol, resulted in slightly poorer
alcohol as a within-subjects factor, and flight
performance than when the Flight A se-
sequence (Flight A versus Flight B on Day 1)
quence was flown on Day 2 with alcohol. Al-
cohol degraded performance in each of the
other three combinations of flight sequence
TABLE 2
and alcohol order.
Average Scores for Overall Pilot Performance, Task Analysis of the multiplicative model scores
Segments, and Performance Aspects Using the also found significantly poorer overall perfor-
Value Additive Model
mance with alcohol, F(1,8) = 9.15, p < 0.02.
Alcohol No Alcohol A significant interaction involving alcohol,
alcohol order, and flight sequence, F(1,8) =
Overall Pilot
Performance 56.89 62.01 15.44, P < 0.01, was also obtained. Here, as
VOR Tracking 62.28 68.40 with the additive model, the only combina-
VC 61.09 66.13
tion of alcohol order and flight sequence that
HC 62.36 69.30
CD 66.85 74.75 resulted in poorer performance under no-al-
Vectoring 64.96 68.56 cohol performance was when the more diffi-
VC 68.78 70.69
cult flight was flown under the no-alcohol
HC 57.05 61.23
CD 73.44 82.08 condition on the first day.
Descent 42.52 47.02 Model-derived performance scores for each
VC 21.09 20.77
task segment were also analyzed. With the
HC 62.36 71.13
CD 68.73 79.69 additive model, significant alcohol perfor-
Collision Avoidance 57.93 64.08 mance decrements were found for both VOR
VC 44.24 48.24
Tracking, F(1,8) = 11.17,p < 0.02, and Colli-
HC 55.52 59.55
CD 79.47 91.33 sion Avoidance, F(1,8) = 6.88, p < 0.05. The
only other significant effects involving alco-
VC = Vertical Control; HC = Horizontal Control; CD = Cockpit
Duties hol involved the Collision Avoidance task
302-June 1988 HUMAN FACTORS

segment, in which an alcohol administration VOR radial. and in Cockpit Duties for the
order effect. F(1,8) = 7.23. P < 0.05, and an Collision Avoidance task segment. in which
Alcohol x Alcohol Order interaction, F(1,8) the primary task should have been execution
= 5.60, p < 0.025. were found. of a vertical evasive maneuver. In both cases
Using the multiplicative model, a signifi- it appears that when alcohol was involved,
cant alcohol main-effect decrement was the pilots were less likely to be attentive to
found only for VOR Tracking, F(1,8) = 18.24, important secondary tasks.
P < 0.005. There was a flight sequence effect Another analysis of the effects of alcohol
for that task segment also, F(1.8) = 7.04, P < upon flight performance was carried out by
0.05, and an Alcohol x Alcohol Order x determining the number of 5-s intervals for
Flight Sequence interaction that was signifi- each of the task segments in which one or
cant for the Collision Avoidance task seg- more of the flight parameters constituting
ment,F(1,8) = 12.26,p < 0.01. As in the other Vertical or Horizontal Control performance
cases in which such an interaction was found, aspects exceeded the "safe" limits defined by
the alcohol effect was related to the difficulty the flight instructors. There were signifi-
level of the flight paired with alcohol and cantly more violations of "safe" vertical con-
whether the alcohol was administered on trol limits during VOR Tracking. F(1,8) =
Day 1 or Day 2. 9.86, P < 0.02, which is consistent with the
Further analyses were carried out in order previous data indicating that the subjects
to determine which performance aspects ac- under the alcohol condition were more likely
counted for the significant main effect of al- to neglect vertical control and, presumably,
cohol found for VOR Tracking. Although per- focus their attention on minimizing CDI
formance for all three aspects was poorer error during this task segment.
under the· alcohol condi tion, as shown in A question of interest is the degree to
Table 2, the only significant difference was in which the findings of the study would have
Vertical Control, F(1,ll) = 9.48, p < 0.02. been different had the multiattribute model
A significant main effect for the Collision of pilot performance not been developed and
Avoidance task segment was also evident. but used in the data analysis. Accordingly, alti-
only with the additive model. Analysis of the tude deviations. CDI errors, and scanning
performance aspect scores for this task seg- and target responses were also analyzed as
ment found a significant main effect for alco- unidimensional performance measures.
hol only for Cockpit Duties, F(1,8) = 10.55, p Altitude root mean square (RMS) error did
< 0.02. Again, Table 2 shows somewhat show a significant alcohol decrement, F(1,8)
poorer performance with alcohol for the = 6.07, p < 0.05, as well as an Alcohol x Al-
other performance aspects, but the differ- cohol Order interaction. F(1.8) = 7.58, p <
ences were not significant. 0.05. Neither the range nor SD of the altitude
It should be noted that in both of these task measures showed an alcohol effect. nor did
segments. the performance aspect that the number of altitude deviations greater
showed a significant alcohol effect was one than 100 ft (25.4 m). The SDs of cm measures
other than the aspect that could be consid- were greater with alcohol, F(1,8) = 5.41, p <
ered the primary flight task. Significant alco- 0.05, although neither RMS nor mean abso-
hol effects were found in Vertical Control for lute errors showed a significant alcohol ef-
the VOR Tracking task segment. in which the fect. Scanning and target response measures
primary task presumably was tracking the analyzed included the duration and number
ALCOHOL AND PILOT PERFORMANCE June 1988-303

of scans when a target did not appear, the re- showing relative neglect of other flight
action time between the appearance of a tar- duties.
get and its detection, the time between target A model such as that used in the present
detection and a significant control input, and study should be of value in any situation in
the average altitude change in response to a which some compensation for the detrimen-
looming target. In all instances the subjects' tal effects of a factor can occur through at-
performance was poorer with alcohol, but tention allocation, and in training situations
the differences were not statistically signifi- in which a good performance score should re-
cant. Five mistakes were made in response to quire continued attention to all flight tasks.
a looming target; there were three cases of Use of such a model does not, of course, pre-
failure to respond, and in two cases the direc- clude examination of specific response mea-
tion of altitude change was incorrect. Four of sures that are of interest. Rather, it provides
the five mistakes were made under the alco- a specific framework within which to exam-
hol condition, and one nonresponse was ine rationally derived components of the pi-
made without alcohol. loting task in a manner that otherwise is not
possible.
DISCUSSION A number of the analyses involving model
scores showed the alcohol variable interact-
It is clear that use of the model provided ing with the order of alcohol presentation
advantages in examining the effects of alco- and the sequence of the two flights. This
hol on simulator flight performance. First, partly reflects the fact that despite the appar-
the model provided a task analysis function ent similarity of the two flights, one was con-
in that the task segments were identified as siderably more difficult. Nevertheless, alco-
important components of the flight scenario hol significantly affected both overall pilot
and thus permitted the examination of alco- performance and VOR Tracking as evaluated
hol's effects on each separately. Further, the by either model and was also a significant
task segments were multidimensional in na- factor for Collision Avoidance in the case of
ture, so that a penalty was incurred if one the additive model. It should be noted that
aspect of flight performance was neglected in these alcohol effects were present with nor-
order to focus attention on another perfor- mal cross-country flight scenarios that did
mance aspect. This characteristic of the mul- not involve unexpected or unusual tasks or
tidimensional model approach is illustrated procedural requirements, when the alcohol
by the analyses of the task segments in which effects might be expected to be most appar-
significant alcohol effects were found. Signif- ent.
icant alcohol effects were found only in Verti- The subjects in this study were relatively
cal Control during the VOR Tracking task low-time pilots, and the question arises as to
segment and in Cockpit Duties during the whether or not experience might mitigate al-
Collision Avoidance task segment, rather cohol's effects on pilot performance. Appar-
than in Horizontal Control and Vertical Con- ently it would be unwise to assume that this
trol, respectively, as might have been ex- would be the case. Even if experience enabled
pected. These findings indicate that under a pilot to compensate to some degree for al-
the alcohol condition, the pilots were concen- cohol's effects, those with more experience
trating on what appeared to be the primary could still be at increased risk given that
task during those segments of the flight while their flying may involve equipment or situa-
304-June 1988 HUMAN FACTORS

tions that impose much greater task de- Billings, C. E., Wick, R. L., Gerke, R. J., and Chase, R. C.
(1972, January). The effects of alcohol on pilot perfor-
mands. It should be pointed out that the Bill- mance during instrument flight (Report No. FAA-AM-
ings et al. (1972) study found that even in a 72-4). Oklahoma City: Civil Aeromedical Institute.
simple aircraft, experienced as well as rela- Henry, P.H., Flueck, J.A., Sanford, J.F., Keiser, H.N.,
McNee, R. C., Walter, W. H., Webster, K. H., Hartman,
tively inexperienced pilots increased their B. 0., and Lancaster, M. C. (1974). Assessment of per-
number of major procedural errors at all BAC formance in a Link-GAT-I flight simulator at three al-
cohol dose levels. Aerospace Medicine, 45, 33-44.
values tested. Alcohol-related accident statis-
Hyman, F. C., Taylor, H. L., Weller, M. H., and Nagel, R. J.
tics (National Transportation Safety Board, (1986, April). Effects of alcohol on flight simulator and
1985) also indicate that experience does not Sternberg memory search task performance. Paper pre-
sented at the 57th Annual Scientific Meeting of the
afford protection from the fatal effects of al- Aerospace Medical Association, Nashville, TN.
cohol. Keeney, R. L., and Raiffa, H. (1976). Decisions with multi-
Decrements in pilot performance such as ple objectives: Preferences and value tradeoffs. New
York: Wiley.
were found in this study may not routinely
Lentz, S. K., and Rundell, O. H. (1976). Sustained control
place an aircraft in imminent risk of an acci- of blood alcohol levels. Alcohol Technical Reports, 5(2),
dent, but the margin of safety would be re- 33-36.
duced, and under circumstances of increased Mundt, J. C., and Ross, L. E. (1987). Multiattribute evalua-
tion of simulator flight performance in research and
demands on the pilot, it is highly likely that training. In Proceedings of the Fourth International
the probability of an accident would be in- Symposium on Aviation Psychology (pp. 151-157). Co-
lumbus: Department of Aviation, Ohio State Univer-
creased significantly. sity.
ACKNOWLEDGMENTS National Transportation Safety Board. (1984). Safety
study.' Statistical review of alcohol-involved aviation ac-
This research was supported by Grant AA6093 from the
cidents (Report No. NTSB/SS84/93). Washington, DC:
National Institute on Alcohol Abuse and Alcoholism. The
Author.
authors thank Hona Lee Harrington for her help in carry-
ing out the study, and Tanya Cunningham and Neil Rob- Ross, L. E., and Mundt, J. C. (1986). Effects of a low blood
inson for serving as flight instructor-consultants during alcohol level on pilot performance. In Proceedings of
development of the multiattribute model. the 30th Annual Meeting of the Human Factors Society
(pp. 1182-1186). Santa Monica, CA: Human Factors
REFERENCES Society.
Aksnes, E. G. (1954). Effects of small dosages of alcohol von Winterfeldt, D., and Edwards, W. (1986). Decision
upon performance in a Link trainer. Journal of Avia- analysis and behavioral research. New York: Cam-
tion Medicine, 25(6), 680-688. bridge University Press.

You might also like