H U M A N F A C T O R S, 1980, 22(6), 725–732

Visual Detection of Driving While Intoxicated

1
DOUGLAS H. HARRIS , Anacapa Sciences, Inc., Santa Barbara, California

Visual cues were identified and procedures were developed to enhance on-the-road detection
of driving while intoxicated (DWI) by police patrol officers. Related research was reviewed;
police officers with demonstrated effectiveness in DWI detection were interviewed; DWI
arrest reports were analyzed; and a study was conducted to determine the frequency of
occurrence and relative discriminability of visual cues. Based on the results, a DWI detection
guide was developed and verified in a field study involving a sample of 10 law enforcement
agencies located throughout the United States. Use of the guide was accompanied by a
statistically significant 12% overall increase in DWI arrest rate.

INTRODUCTION inhibiting enforcement motivation (Arthur

Young and Company, 1974; Oates, 1974)
On-the-road detection of driving while in-
might explain the low arrest rates. However,
toxicated (DWI) requires the observation and
even when persons DWI have been observed
interpretation of visual cues by police patrol
by patrol officers who were highly motivated
officers. The effectiveness of DWI detection,
to arrest, the arrest rate was found to be very
therefore, is a function of the frequency with
low (Beital, Sharp, and Glauz, 1975).
which patrol officers see and recognize cues
As determined from roadside breath-
indicative of DWI and the extent to which ob-
testing surveys conducted throughout the
served cues discriminate between DWI and
United States (Lehman, Wolfe, and Kay,
driving while sober (DWS). What cues occur
1975), about 6% of people driving at night
frequently enough to be useful? Which cues
have a blood alcohol concentration (BAC)
discriminate most accurately between DWI
equal to or greater than 0.10% by weight, the
and DWS? This research was conducted to
legal definition of DWI in most cities. About
answer these and related questions and to
15% have a BAC level equal to or greater than
provide patrol officers with a practical guide
0.05. Thus, if DWI were defined at the BAC ≥
for DWI detection.
0.10 level, the probability of detecting DWI
The Detection Problem from a random stop would be 0.06; at BAC ≥
0.05, the probability would be 0.15. Visual
Only a very small proportion of persons cues that aid in discriminating between DWI
DWI are arrested for this offense—only about and DWS can serve to increase detection
one in 2000 (Summers and Harris, 1978). probabilities above these chance levels. Thus,
Limited enforcement resources and factors the key to enhanced DWI detection is deter-
1
mination of the relative discriminability of
Requests for reprints should be sent to Dr. Douglas H.
Harris, Anacapa Sciences, Inc., P.O. Drawer Q, Santa Bar-
visual cues that are likely to be observed in
bara, CA 93102. association with DWI.
c 1980, The Human Factors Society, Inc. All rights reserved.


726–December, 1980 HUMAN FACTORS

Related Research provided the basis for a preliminary listing of

129 visual cues potentially useful for DWI de-
Numerous studies have investigated the
tection.
effect of alcohol on driving behavior; they
have employed laboratory apparatus, driving ON-THE-ROAD DETECTION STUDY
simulators, and instrumented vehicles in the
field. Reviews of many of these studies have Approach
been provided by Heimstra and Struckman An on-the-road study of DWI detection was
(1973) and Perrine (1974 and 1975). The re- conducted to determine the relative dis-
sults, however, have been relevant to the cur- criminability and frequency of occurrence of
rent project only indirectly. Although sub- visual detection cues under conditions typi-
stantial evidence has been developed to indi- cally encountered by patrol officers. Trained
cate that alcohol-induced driver impair- observers accompanied police officers on pa-
ment is exhibited mainly in four driving trol and recorded instances of driving be-
functions—steering control, velocity control, havior and vehicle actions that deviated from
time-sharing of attention, and information normal. In each instance, the police officer
processing—the findings have not been spe- stopped the vehicle and, if possible, measured
cific enough to permit the identification and the BAC of the driver with a portable breath
assessment of visual detection cues. On the tester. Measures of BAC were obtained in 93%
other hand, these findings provided a useful of all stops. In addition to cue descriptions
framework for the collection and analysis of and BAC level, the observer recorded driver
additional data. characteristics and the circumstances and
Lists of cues have been developed previ- conditions under which the stop was made.
ously through interviews with police officers Since the data-collection effort required con-
experienced in DWI detection (Carnahan, ducting prearrest breath tests of drivers, the
Holmes, Keyes, Stemler, and Dreveskracht, study was conducted in two states, Indiana
1974); a listing was also developed as part of and North Carolina, which permitted, by
the current study. The resulting listings have statute, prearrest breath testing.
been both comprehensive and logically orga-
nized; however, lists of this type can have Results
only limited utility for DWI detection without
A total of 643 DWI detection events was ob-
associated information about relative fre-
served and recorded. The sample of detection
quencies of cue occurrence or relative cue
events was nearly identical to the national
discriminability.
sample of arrest reports, relative to the dis-
tributions of times at which stops were made,
ANALYSIS OF DWI ARREST REPORTS
the location of stops (urban or rural), and the
Initially, an analysis was completed of a percentages of male and female drivers
sample of 1288 DWI arrest reports from nine stopped. A total of 1681 cue occurrences was
different police agencies throughout the recorded during the 643 detection events, an
United States. A total of 3658 visual detection average of 2.6 cues per event, which is close to
cues was reported in the sample, an average the average of 2.8 cues per arrest reported in
of 2.8 cues per arrest. Frequency distributions the national sample of arrest reports. These
prepared from the data, combined with the results suggest that the stops made in the de-
results of previous research and cue listings tection study were typical of those made
obtained from experienced patrol officers, normally by police patrols.
DOUGLAS H. HARRIS December, 1980–727

In the detection study, 39% of the drivers counted for 92% of the 1681 cue occurrences
stopped had a BAC less than 0.05; 23% had a in the detection study. The 23 resulting cues
BAC greater than 0.05 but less than 0.10; and are listed in Table 1; frequency of occurrence,
38% had a BAC equal to or greater than 0.10. P(BAC ≥ 0.10), and P(BAC ≥ 0.05) are pre-
Of course, most of the national sample of DWI sented for each cue.
arrests, 96%, reported drivers with BAC equal A listing of the 134 cues observed in the
to or greater than 0.10. detection study served as the starting point
Cues were combined and redefined, ulti- for the analysis. This listing included 118 cues
mately, into a set of 23 visual cues that ac- from the preliminary listing of 129 cues, plus

TABLE 1
DWI Visual Detection Cues Derived from On-the-Road Study
Occurrence (Times
Visual Cue in 1000 Detections) P(BAC ≥ 0.10) P(BAC ≥ 0.05)
Stopping (without cause) in 29 0.69 0.90
traffic lane
Following too closely 29 0.62 0.76
Turning with wide radius 35 0.60 0.83
Appearing to be drunk 57 0.58 0.75
Driving on other than 42 0.57 0.79
designated roadway
Straddling center or lane 65 0.57 0.78
marker
Almost striking object or 62 0.56 0.71
vehicle
Slow response to traffic 20 0.50 0.55
signals
Headlights off (at night) 27 0.48 0.67
Signaling inconsistent with 49 0.47 0.71
driving actions
Weaving 145 0.47 0.69
Tires on center or lane 101 0.47 0.67
marker
Drifting 108 0.46 0.70
Swerving 49 0.45 0.73
Accelerating or decelerating, 81 0.44 0.67
rapidly
Slow speed—more than 10 MPH 32 0.44 0.66
below limit
Fast speed—more than 10 MPH 101 0.37 0.55
above limit
Failing to respond to 85 0.36 0.53
traffic signals or signs
Braking erratically 23 0.35 0.74
Stopping inappropriately 33 0.33 0.61
other than in lane
Turning abruptly or 48 0.31 0.58
illegally
Driving into opposing or 37 0.30 0.54
crossing traffic
Driving with vehicle 42 0.29 0.43
defect(s)
728–December, 1980 HUMAN FACTORS

16 new cues observed during the detection age of the driver. All correlations were statis-
study. Since this initial listing was empiri- tically significant ( p < 0.01).
cally derived, it was not logically cohesive The one variable that most influenced the
and contained many infrequently occurring frequency of cues observed was patrol em-
cues. Therefore, cues were redefined to phasis. About 58% of the detection events oc-
maximize frequency of occurrence, maintain curred under general patrol, in which DWI
levels of cue discriminability, and enhance was just one of many possible offenses of con-
cue understandability and applicability. For cern to the patrol officer. About 42% of the
example, the cue “weaving” was defined as detection events occurred under patrols
the deviation from vehicle path alternately which emphasized DWI enforcement. The
toward one side of the roadway and then the correlation between cue frequency distribu-
other, creating a zig-zag course. The original tions obtained under the two types of patrol
listing contained seven weaving cues that was only 0.22.
differed from each other mainly as a function
of weave amplitude; i.e., weaving in lane, lane Conclusions
to lane, lane to shoulder, across lane, across (1) Although the potential number of visual de-
centerline, center of roadway with no cen- tection cues is large, most detection events
can be accounted for by a relatively small
terline, shoulder to shoulder. At little loss in number of cues.
cue discriminability and at substantial in- (2) There are large differences among visual de-
crease in cue frequency of occurrence and tection cues in the frequency with which they
occur with DWI and in their ability to distin-
applicability, all seven were combined into guish between DWI and DWS.
the single weaving cue. (3) Typically, a visual detection cue is observed
Typically a cue occurred with one or more with one or more other cues. However, there
are no subsets of specific cues that occur fre-
other cues, the average number of cues per quently together.
detection event being 2.6. Two or more cues (4) Except for patrol emphasis, the conditions
were observed in 66% of the detection events. under which cues are observed have relatively
little influence on frequency of cue occur-
However, relatively few subsets of cues oc- rence.
curred together consistently. Only 9 subsets (5) The basis exists for the development and
of the 134 cues occurred 10 times or more in evaluation of a DWI detection guide to facili-
tate the application of research findings to
the sample of 643 detection events, and essen- on-the-road detection of DWI by police patrol
tially no subset of the 23 cues in the final set officers.
occurred frequently enough to warrant
DWI DETECTION GUIDE
further consideration.
Relatively high correlations (ranging from The extent of competing demands placed
0.62 to 0.82) among distributions of cue fre- upon patrol officers—the variety of situations
quencies were obtained under alternative de- likely to be encountered, the stringent de-
tection conditions: duration of observation, mands on available time, the need for rapid
distance at which the cue was observed, time response, and the large amount of other in-
of day, lighting conditions, location (urban or formation that must also be learned and
rural), condition of the vehicle, sex of the retained—suggested that the findings of this
driver, and number of passengers in the vehi- study be presented for use simply and di-
cle. More modest correlations (ranging from rectly. Therefore, a DWI detection guide was
0.49 to 0.56) were obtained for the follow- developed to transform the research findings
ing conditions: number of traffic lanes, divided into a practical aid for DWI detection. Be-
vs. undivided highway, traffic density, and cause the empirical results were not neces-
DOUGLAS H. HARRIS December, 1980–729

sarily simple or free of subtlety, extrapolation guide were obtained by rounding the values
and judgment were exercised during this pro- shown in Table 1. The special adjustments for
cess. Guide development was governed by the multiple cues were obtained by linear regres-
following criteria: account for the largest sion from average DWI probability values
number of detection events with the smallest computed for the following conditions of cue
number of detection cues; enhance the dis- occurrence: cue observed as one of one or
criminability of available detection cues; more cues; cue observed as one of two or
employ a probabilistic output; accommodate more cues; and cue observed as one of three
multiple cue occurrences; accommodate al- or more cues. Values in the guide are those
ternative enforcement statutes and policies; computed for cues when they were observed
and emphasize simplicity, practicality, and as one of one or more cues. On the average,
ease of use. these values were found to be increased by 5
The resulting Drunk Driver Detection where the cue was observed as one of two or
Guide is presented in Figure 1. The guide, to- more cues and increased by 10 when a cue
gether with a booklet of cue definitions, was was one of three or more cues. Similarly,
designed as a simple performance aid that under each condition, probability values for
could be implemented by patrol officers after P(BAC ≥ 0.05) were found to be increased by
a brief training session. 20 over the value for P(BAC ≥ 0.10).
The probability values contained in the
FIELD VERIFICATION
Approach
A field test of the guide was conducted with
DRUNK DRIVER DETECTION GUIDE a sample of 10 law enforcement agencies at
Percentage of nightime drivers with BAC equal to or greater than .10
locations throughout the United States. The
Visual Cues 10 agencies employed different types of pa-
STOPPING [WITHOUT CAUSE] IN TRAFFIC LANE ........ 70 trols: general patrols responsible for both
FOLLOWING TOO CLOSELY .......................................... 60
TURNING WITH WIDE RADIUS ..................................... 60 criminal and traffic enforcement; general
APPEARING TO BE DRUNK ........................................... 60 traffic patrols responsible for enforcement of
DRIVING ON OTHER THAN DESIGNATED ROADWAY ... 55
STRADDLING CENTER OR LANE MARKER ................... 55 traffic laws; selective traffic patrol responsi-
ALMOST STRIKING OBJECT OR VEHICLE .................... 55 ble mainly for DWI enforcement. The test
SLOW RESPONSE TO TRAFFIC SIGNALS ..................... 50
HEADLIGHTS OFF [AT NIGHT] ...................................... 50 employed a within-subjects type of experi-
SIGNALLING INCONSISTENT WITH DRIVING ACTIONS 45 mental design and several measures likely to
WEAVING ...................................................................... 45
TIRES ON CENTER OR LANE MARKER ......................... 45 reflect the impact and utility of using the
DRIFTING ....................................................................... 45 guide and to verify the detection probabilities
SWERVING ..................................................................... 45
ACCELERATING OR DECELERATING RAPIDLY .................... 45 contained in the guide.
SLOW SPEED [MORE THAN 10 MPH BELOW LIMIT] ............ 45 Measures related to DWI enforcement ef-
FAST SPEED [MORE THAN 10 MPH ABOVE LIMIT] .............. 35
FAILING TO RESPOND TO TRAFFIC SIGNALS OR SIGNS ....... 35 fectiveness were obtained from each agency
BRAKING ERRATICALLY .................................................... 35 during a 12-mo baseline period and during a
STOPPING INAPPROPRIATELY [OTHER THAN IN LANE] ........ 35
TURNING ABRUPTLY OR ILLEGALLY .................................. 30 3-mo test period in which the guide was used.
DRIVING INTO OPPOSING OR CROSSING TRAFFIC .............. 30 Three measures were obtained monthly for
DRIVING WITH VEHICLE DEFECT[S] ................................... 30
Special Adjustments to the Percentages
each agency during both baseline and test pe-
• 2 cues: Add 5 to the larger percentage. riods: number of DWI arrests per 100
• 3 or more cues: Add 10 to the largest percentage.
• To predict BAC equal to or greater than .06: Add 20 to the percentage obtained person-hours of patrol, frequencies with
for drivers with BAC equal to or greater than .10.
which detection cues were reported on arrest
Figure 1. DWI detection guide. reports, and BAC levels of persons arrested.
730–December, 1980 HUMAN FACTORS

1.0
on any of three measures that might have re-
DWI ARRESTS PER 100 PERSON-HOURS OF PATROL

0.9 flected changes in DWI detection practices.

The number of visual detection cues recorded
0.8
.74 per arrest did not increase; a shift to the use
0.7
.65 of higher probability cues was not statisti-
0.6
cally significant; and the BAC levels of per-
sons arrested did not decrease significantly.
0.5
Probability values contained in the guide
0.4 were verified by the proportions of drivers
apprehended during the field test who were
0.3
found to have BAC levels equal to or greater
0.2 than 0.05 and 0.10. Average guide probability
0.1
values over all cues were essentially the same
as the average probabilities calculated from
BASELINE TEST field-test data.
Figure 2. Average DWI arrest rates for all 10 par- Correlations between guide DWI prob-
ticipating agencies during baseline and test periods. abilities and test probabilities for individual
cues ranged from 0.48 to 0.59 and were statis-
During the 3-mo test period, two additional tically significant ( p < 0.05) in all cases.
measures were obtained: ratio of drivers DWI Comparisons of guide and test values must be
to drivers apprehended for each cue or cue made in light of two important factors that
combination and opinions and suggestions of would be expected to decrease the correla-
participating police officers regarding use of tions. First, the probabilities contained in the
the guide. guide were derived from data collected using
procedures that were substantially different
Results
from those employed for collecting data in
Collectively, for all 10 participating agen- the field test. Guide probabilities were based
cies, DWI arrest rate was 12% higher during on data obtained by stopping each driver ob-
the test period than during the baseline pe- served to be exhibiting deviant driving be-
riod. This difference was statistically signifi- havior and administering a breath test to the
cant ( p < 0.01). Comparison of baseline and driver. Observers accompanied patrol officers
test periods is shown graphically in Figure 2. for purposes of recording the data. Field-test
Individually, five agencies had significant in- probabilities, on the other hand, were ob-
creases in DWI arrest rates of up to 94%; four tained from data recorded on special forms
agencies did not change significantly; and during regular patrol by the patrol officers
one agency had a significant decrease. themselves. Some detection procedures, such
Changes within an agency were consistent as the use of radar to detect fast speed, dif-
among the different types of patrols em- fered substantially from those used in the
ployed by the agency. These results were earlier study. Each time a driver was ap-
based on a total of 5348 arrests made during prehended, the officer used normal proce-
the 15-mo period by 466 patrol officers during dures, such as field sobriety tests, to estimate
788 200 person-hours of patrol activity. the three categories of BAC. Verification, by
There were no statistically significant dif- followup chemical tests, of about two-thirds of
ferences between baseline and test periods the BAC ≥ 0.10 estimates showed them to be
DOUGLAS H. HARRIS December, 1980–731

99% accurate. The procedures used in the with cues contained in the guide were verified
field test were dictated by the need to ob- by field-test results, providing a basis for
tain a large, geographically representative using guide values with confidence. Although
sample. some modifications to guide values were in-
Second, guide probability values used for dicated, the overall result was one of verify-
the correlations were obtained directly from ing the average probability values as well as
the guide rather than from the original data. the values associated with individual cues. By
Thus, they were rounded-off values for P(BAC combining field-test and detection study
≥ 0.10) that were extended to multiple cue data, the guide values will be based on a total
conditions and to P(BAC ≥ 0.05) through ap- of 4662 detection events that occurred at 12
plication of the special adjustments present- different locations throughout the United
ed at the bottom of the guide. Correlations States.
between field-test values and the actual val- Some difficulty might be expected in gain-
ues obtained from the earlier detection study ing acceptance of the guide by police officers
were higher in all cases, ranging from 0.49 to experienced in DWI enforcement. Many feel
0.68. they have little or nothing to learn from the
Experienced police officers who used the guide or that detection is not a primary
guide were generally skeptical that use of the problem in DWI enforcement. On the other
guide would enhance their own DWI detec- hand, after using it, officers stated that the
tion ability. However, most officers consid- guide would be of value for a variety of pur-
ered the guide to be a valuable aid for in- poses.
creasing patrol awareness of useful detection
cues, training inexperienced patrol officers, ACKNOWLEDGMENTS
preparing DWI arrest reports, and supporting
The research was conducted under Contract DOT-HS-
court testimony. 7-01538 with the National Highway Traffic Safety Admin-
istration, Department of Transportation. The following
CONCLUSIONS members of the Anacapa Sciences, Inc., staff contributed
significantly to the project: Steven M. Casey, Robert A.
The utility of the Drunk Driver Detection Dick, James B. Howlett, Christopher J. Jarosz, and R.
Glen Ridgeway. The opinions, findings, and conclusions
Guide developed for on-the-road detection of expressed are those of the author and not necessarily those
DWI was demonstrated. Use of the guide re- of the National Highway Traffic Safety Administration.
sulted in an overall increase in DWI arrest
rate of 12%. This increase took place in a sam- REFERENCES
ple that included 10 police agencies located Arthur Young and Company. Factors influencing alcohol
throughout the United States. These agencies safety action project police officer’s DWI (driving while
employed various types of patrols, included a intoxicated) arrests. Washington, DC; Department of
Transportation, NHTSA, Report No. DOT HS-801-151,
wide range of geographic and traffic condi- April, 1974. (NTIS No. PB 232 538)
tions, and reflected different levels of motiva- Beital, G. A., Sharp, M. C., and Glauz, W. D. Probability of
arrest while driving under the influence of alcohol.
tion for DWI enforcement. Although there Journal of Studies on Alcohol, 1975, 36, 109–116.
were no statistically significant changes in Carnahan, J. E., Holmes, D. M., Keyes, J. A., Stemler,
J. D., and Dreveskracht, C. L. DWI law enforcement
detection practices, such as those revealed by training project, student manual. East Lansing, MI:
greater use of the more discriminating cues or Michigan State University, Contract DOT HS-334-3-
645 with NHTSA, May, 1974.
by arrests of more drivers with lower BAC Heimstra, M. W., and Struckman, D. L. The effects of al-
cohol on performance in driving simulators. Vermil-
levels, trends were in the expected directions. lion, SD: University of South Dakota. Human Factors
The DWI probability values associated Laboratory, 1973.
732–December, 1980 HUMAN FACTORS

Lehman, R. J., Wolfe, A. C., and Kay, R. D. A computer National Safety Council, Committee on Alcohol and
archive of ASAP roadside breath-testing surveys 1970– Drugs, 1974.
1974. Ann Arbor, MI: Highway Safety Research Insti- Perrine, M. W. Alcohol experiments on driving-related be-
tute, NHTSA Report No. DOT HS-801 502. January, havior: A review of the 1974 literature. Chicago, IL:
1975. (NTIS No. PB 242 074) National Safety Council, Committee on Alcohol and
Oates, J. F., Jr. Factors influencing arrests for alcohol- Drugs, 1975.
related traffic violations. Darien, CT: Dunlap and Summers, L. G., and Harris, D. H. The general deterrence
Associates, NHTSA Report No. DOT HS-801 230, of driving while intoxicated (Volume I) system analysis
September, 1974. (NTIS No. PB 237 004) and computer-based simulation. Santa Barbara, CA:
Perrine, M. W. Alcohol experiments on driving-related be- Anacapa Sciences, Inc., Contract DOT HS-6-01456
havior: A review of the 1972–73 literature. Chicago,IL: with NHTSA, January, 1978.