Epidemiology and Data Management

 

Epidemiologic Perspectives on
Drunk Driving

M.W. Perrine
Vermont Alcohol Research Center, Burlington, Vermont

Raymond C. Peck

Research and Development Office
California Department of Motor Vehicles, Sacramento, California

James C. Fell

National Center for Statistics and Analysis, National Highway Traffic Safety
. Administration, Washington, D.C.

What is the one element found in approximately half the U.S. highway fatalities? This
question has been raised over the last few decades and the answer is still the same:
Alcohol. This answer generates another question: If a single, identifiable element is
involved in such a large portion of a serious public health and public safety problem,
should it not receive top priority for investigation, intervention, and prevention?

Alcohol produces both pleasure and pain, euphoria and depression. Alcohol also
produces many jobs and billions of dollars in tax revenues to the States and to the Nation.
Each year, alcohol also produces unintentional death to thousands and injury to millions.
When mixed with driving, alcohol is the basis for a major public health and public safety
problem. In our automotive society, the car is used for almost all facets of social activity.
Therefore, since alcohol is involved in many aspects of social behavior, driving after
drinking is a relatively frequent occurrence. Fortunately, the vast majority of such
driving-after-drinking instances do not result in crashes. One very important task for
researchers is to identify variables that differentiate between those driving-after-drink-
ing instances that do result in a crash and those that do not.

How do we learn about the contribution of alcohol to unintentional injury and death
on the highways? In attempting to do so, we still find a large gap between description
and explanation that, at this time, can be bridged only provisionally through inference.
Two widely separated research approaches have been used to date asa basis for inferring
the contribution of alcohol to highway crashes: epidemiologic and experimental.

 

NOTE: Primary responsibility for parts of this chapter is as follows: J.C. Fell, the subsection entitled “Aleohol
Involvement in Fatal Highway Crashes”; R.C. Peck, the subsection entitled “Charactersitics of Drunk
Drivers”, MW, Perrine, all other sections. The first author is grateful to Robert B. Voas for early discussions
of this chapter and for material provided in the sections on “Other Roadside Research” and “Enforcement
_ Checkpoints.” Preparation of M.W. Perrine’s part of this chapter and production of the manuscript were

Supported by PHS Research Grants AA74, AA06926, and AA07876 from the National Institute on Alcohol
Abuse and Alcoholism. .

35
36

BACKGROUND PAPERS

The epidemiology of alcohol and highway safety can be traced from the first review
of the problem presented in 1933 (Miles 1934). Over the years, high blood alcohol

concentration (BAC) has been thoroughly implicated in serious and fatal injury highway

crashes by post hoc epidemiologic studies. Most evidence for relating-this alcohol
contribution to highway crashes has been obtained by examination of the distribution of
BAC both among drivers involved in actual crashes (fatal and nonfatal) and—on the
basis of case-control roadside surveys — among drivers using the highways, but not
involved in crashes at the time. A number of such casé-control studies have demonstrated
that alcohol is overrepresented among deceased drivers relative to drivers in the popula-
tion-at-risk using the highways at correspon ing times and places (e.g., Borkenstein et
al. 1964, 1974; Perrine et al. 1971; for reviews see NRC 1987; NHTSA 1985; Perrine
1975a, b).

The second approach consists of controlled administration of alcohol in experiments:
conducted on isolated variables that are assumed to be relevant for actual driving.
Alcohol impairment of real-world driving performance is then typically inferred from
the mosaic of these bits of behavior examined separately in the laboratory, in driving
simulators, and in instrumented cars driven on closed courses (NRC 1987).

Both research approaches to the study of drunk driving are necessary and have been
productive (NRC 1987, Perrine 1976). However, this chapter is limited to epidemiologic
aspects; it is organized as follows:

A discussion of the scope of the drunk-driving problem from an
"epidemiologic perspective

A brief outline of the major components involved in studying the problem
by means of the available data sources

A review of the most relevant literature, focusing on alcohol involvement
in fatal as well as nonfatal highway crashes and in noncrash drivers; crash
risk and alcohol; and characteristics of drunk drivers

An examination of current issues and problems
Recommendations

Scope of the Problem

The primary problem clearly consists of ‘those motor vehicle crashes that result in fatal
injuries. It is now generally well established that alcohol is involved in approximately half
of all such fatal crashes. For example, the total number of highway fatalities in 1986 was
46,056, of which some 24,000 (52 percent) involved alcohol. More specifically, BACs
exceeded the typical legal limit (0.10) in 41 percent of all fatal crashes.

Anestimated 4.8 percent of deaths in the United States during 1980 were directly or
indirectly attributable to alcohol (NIAAA 1987). Of these, motor vehicle crashes were
the largest single cause of death. Approximately 26,000 deaths in 1980 were attributed
to alcohol in motor vehicle crashes; these deaths constituted about 27 percent of the total
number of deaths (approximately 98,000) attributable to alcohol (NIAAA 1987, p.6).
The number of alcohol-involved motor vehicle deaths is about two times that of the
second largest single cause of alcohol-involved death, namely, homicide (approximately ~
12,000 or 12 percent) (NIAAA 1987, p.6).

The scope of the alcohol and traffic safety problem has recently been reviewed briefly
both from a public health perspective (NIAAA 1982, 1985, 1987) and from a public safety
perspective (NHTSA 1985, 1987b; NRC 1987). In this chapter, the problem is examined

further to provide a more integrated synthesis of the literature from both these
perspectives.
EPIDEMIOLOGY AND DATA MANAGEMENT

Major Components of the Study

Aside from epidemiologic methodology considerations, the major components in-
volved in the present perspective on drunk driving consist of the data sources. The two
primary sources of data for this area are official records and surveys of various types.

The official records consist primarily of the following:
The citation report for driving under the influence of alcohol (DUT)
The accident report, if alcohol is involved
The prosecutor record
Court records
Department of Motor Vehicle records
Treatment/service provider records
Probation department records

Of special importance are those reports of accidents in which a fatality resulted, since
these data are collected at the State level and then forwarded to the Fatal Accident
Reporting System (FARS) of the National Highway Traffic Safety Administration
(NHTSA). As an example of using such data, analyses of DUI processing from the point
of the arrest citation through the other official records, including the postconviction

countermeasures, have been prepared for the State of California (Perrine 1984;
Helander 1986; Peck 1987).

The other major source of data for epidemiologic studies consists of surveys. The
main varieties are:

roadside surveys,

telephone surveys (in recent years, the random- digit-dialing telephone
survey),

household surveys, and
special location surveys (bars, jails, etc.)-

Of these various types, only the roadside surveys can obtain direct measurements of
the major criterion variable — namely, BAC — from drivers actually using the roads at the
time. All the other survey methods depend on self-reported information from the
respondents, including data concerning driving after drinking. Thus, only the direct
measurement of BAC at roadside can be used to provide criterion measures for
estimating alcohol crash risk and for evaluating the impact of countermeasure programs
on the motoring public.

Review of the Most Relevant Literature

Alcohol Involvement in Fatal Highway Crashes

In 1986, 46,056 people were killed in traffic crashes (NHTSA 1988), which are the
leading cause of death for Americans age 6-34 (Richardson 1985). Traffic fatalities in
1986 resulted in 1,425,517 years of potential life lost before age 65, an amount greater
than deaths from cancer, heart disease, and all other causes. Traffic crashes cost society
approximately $74 billion annually in terms of damage, insurance costs, injury treat-
ments, lost work, and so forth. (NHTSA 1987c). Since 1900, over 2,600,000 Americans
have died in traffic crashes; that is 1,500,000 more than the total number of Americans
killed in all the wars in U'S. history.

37
38

BACKGROUND PAPERS

It is well known that alcohol is a leading factor in traffic crashes. It was involved in
over half of the traffic fatalities in 1986, resulting in close to 24,000 deaths (NHTSA 1988).
Each year, nearly 560,000 additional people suffer injuries in alcohol-related crashes —

are serious (NHTSA 1988).

- an average of one person injured every minute of the day. About 43,000 of these injuries

During the 1982-86 period, approximately 119,000 people lost their lives in alcohol-
related traffic crashes — an average of one alcohol-related fatality every 22 minutes over
the past 5 years. About two out of five Americans will be involved in an alcohol-related
crash in their lifetime (NHTSA 1987a). Approximately 1,800,000 drivers were arrested
in 1986 for DUI—an arrest rate of about 1 out of every 90 licensed drivers in the United -
States (Greenfield 1988).

The problem is especially devastating for young people. In 1986, more than 40 percent
of all teenage deaths resulted from motor vehicle crashes. Over half of these were alcohol
related, making alcohol-related traffic crashes the leading cause of death for teenagers.
For traffic crash victims age 20-24, close to 70 percent of the 8,000 who died in 1986 were
in alcohol-related crashes (NHTSA 1988). The probability that a given death is due to
a traffic fatality is 55 times as great for a 20-year-old male as for a 65-year-old male; the
corresponding ratio for females is 43 (Evans 1987).

The average BAC of drinking drivers involved in fatal crashes was 0.15 in 1986
(NHTSA 1985). The legal intoxication limit in most States is 0.10. In a recent survey of
drivers jailed for drunk driving offenses, over a quarter of the drivers had consumed at
least 20 beers or 13 mixed drinks within 3-4 hours before they were arrested (Greenfield
1988). Research has shown that a driver with a BAC of 0.15 has a 26-times greater
probability of being involved in a crash than a sober driver (NHTSA 1985).

FARS indicated that 41 percent of the traffic fatalities in 1986 involved either a driver
or a pedestrian with a BAC of 0.10 or greater. This percentage translated to 18,890
fatalities. An additional 11 percent (5,100 fatalities) involved a driver or pedestrian with
some alcohol (BAC =0.01-0.09). Only 48 percent of the fatalities involved all drivers
and pedestrians with zero alcohol.

Alcohol involvement did vary by time of day, day of week, and type of crash (table 1).
Seventy-seven percent of the fatal crashes that occurred between 8 p.m. and 4 a.m. on
any night of the week involved alcohol. Alcohol was also much more prevalent in
single-vehicle crashes than multiple-vehicle crashes. Almost half the collisions resulting

Table 1. Alcohol involvement in fatal crashes: 1986

 

 

 

BAC

0.00 0.01-0.09 0.10 and higher

Crashes N (percent) (percent) (percent) —
Fatal 41,062 48 11 41
Daytime (4 a.m. - 8 p.m.) - 23,828 67 9 25
Nighttime (8 p.m. - 4 a.m.) 16,900 23 14 63
Weekday 22,700 59 9 32
Weekend (8 a.m. Fri- 4a.m.Mon) 16,277 35 12 53
Single vehicle 17,114 38 11 51
 Multivehicle 16,244 58 11° 31
Nonoccupant 7,004 = 51 9 40

(pedestrian/bicyclist)

 
EPIDEMIOLOGY AND DATA MANAGEMENT

Table 2. Drivers and nonoccupants (pedestrians/bicyclists) involved in fatal
. crashes: 1986 -

 

 

 

BAC
0.00 0.01-0.09 0.10 and higher
N (percent) (percent) (percent)
All drivers , 60,297 66 8 26
Driver fatalities 26,613 52 9 39
Surviving drivers 33,684 77 8 15
Nonoccupant fatalities 7,770 64 7 29
Male drivers 46,622 63 9 28
Female drivers 12,734 79 6 15

 

in anonoccupant (pedestrian or pedalcyclist) death involved alcohol, mostly on the part
of the pedestrian.

When examining data for all drivers involved in fatal crashes, keep in mind that in
multiple-vehicle crashes, at least two drivers are involved in one crash. In 1986, 60,297
drivers were involved in the 41,067 fatal crashes. Twenty-six percent of these drivers were
legally intoxicated (BAC greater than or equal to 0.10) at the time of their crashes (table
2). Of the 26,613 drivers who were killed in their crashes, 39 percent were legally
intoxicated compared with only 15 percent of the drivers who survived fatal crashes. Male
drivers were almost twice as likely to have over 0.10 BAC at the time of their crashes as
female drivers (28 percent versus 15 percent).

Alcohol involvement did vary substantially by driver age in 1986 (table 3). While 21
percent of teenage drivers were legally intoxicated at the time of the crash, an additional
13 percent had also been drinking. Drivers 20-24 years old had the highest alcohol
involvement rate: 47 percent. In contrast, only 7 percent of drivers age 65 and older were
legally intoxicated at the time of their crash.

Examining certain combinations revealed that while almost two-thirds of the fatal

Table 3. Alcohol involvement by driver age, 1986

 

 

Driver’s BAC
Driver’s age 0.00 0.01-.09 0.10 and higher

(percent) . (percent) (percent) N*
16-19 66 13 21 7,854
20-24 53 12 35 11,427
25-34 59 8 33 16,163
35-54 72 6 22 14,305
55-64 81 5 14 4,017
65 and older 89 4 7 4,881
All ages 66 8 26 60,297

 

*N = Number of drivers in age group where age was known.
40

 

 

 

BACKGROUND PAPERS
Table 4. Drivers involved in fatal crashes, 1986
BAC _
0.00 0.01-.09 0.10 and higher
N ' (percent) (percent) (percent)
Male/weekend/night 10,573 38 14 49
Female/weekday/day 6,503 90 3 7
Driver age groups

15-20 10,467 64 3 23
21-44 34,518 60 9 31
82 5. 13

45 and older 13,968

 

crashes involving a male driver ona weekend night were alcohol related, only 10 percent
of the crashes involving a female driver in a weekday crash in the daytime involved alcohol
(table 4).

Alcohol involvement was also found to vary considerably by the type of vehicle driven

(indicating the type of driver, in most cases) (table 5). Drivers of motorcycles involved

in fatal crashes had by far the highest alcohol involvement rate: 54 percent. Only3 percent
of heavy-truck drivers involved in fatal crashes had BACs over 0.10. Drivers of older
vehicles were more often legally intoxicated than drivers of newer vehicles (34 percent
versus 22 percent).

Intoxicated drivers in fatal crashes also tended not to use safety belts. Of the fatally
injured drivers who were at zero alcohol, 20 percent were wearing safety belts compared
with only 7 percent of ‘the fatally injured drunk drivers. Thirty-six percent of the
zero-alcohol surviving drivers were reported as using belts, in contrast to only 15 percent
of the intoxicated surviving drivers.

Contrary to some popular misconceptions, the victims of alcohol-related fatal crashes
are most often the drinking driver or drinking pedestrian. Two-thirds (66 percent) of the
23,990 victims of alcohol-related crashes in 1986 were the drinking driver or drinking
pedestrian (table 6). An additional 20 percent of the victims were passengers in the

Table 5. Drivers involved in fatal crashes, 1986

 

BAC

Drivers of: 0.00 0.01-0.09 0.10 and higher
N (percent) (percent) (percent)

Motorcycles 4,542 46 B 41
Passenger cars 35,920 65 9 26
Light trucks and vans 11,724 63 8 29
Medium trucks 653 92 3 5
Heavy trucks 4,355 95 2 3
Older vehicles (older than 1976) 13,168 59 9 34
Newer vehicles (1984-87) 15,579 70 8 22

 
EPIDEMIOLOGY AND DATA MANAGEMENT

Table 6. Alcohol-involved fatal crash victims, 1986

—

 

 

N Percent
Drinking drivers killed 13,190 55
Drinking pedestrians and pedalcyclists killed 2,640 11
Passengers in drinking driver’s vehicle killed 4,800 ' 20
Sober drivers killed in crash with drinking driver’s vehicle 1,680 7
Passengers in sober driver’s vehicle killed in crash
with drinking driver 960 4
Sober pedestrians/pedalcyclists killed by drinking driver - 720 3
Total 23,990 100

 

drinking driver’s vehicle. Additional analyses revealed that in fatal crashes where BACs
were known for drivers and their passengers, 36 percent of the time the driver was legally
intoxicated but the passenger was not.

Table 7 shows the basic trend with regard to the alcohol problem in fatal crashes over
the past 5 years. The percentage of drivers in fatal crashes who were intoxicated
(BAC =0.10 or greater) at the time of the crash decreased from 30 percent in 1982 to 26
percent in 1986—a 13-percent reduction, which is substantial. The reduction was
especially great for teenage drivers (table 8). While 29 percent of the teenage drivers in
1982 were legally intoxicated, this amount dropped to 21 percent in 1986, a 28-percent
reduction. While this teenage driver trend is encouraging, one must still keep in mind
that teenage driver involvement in fatal crashes per mile driven is substantially higher
than other driver age groups (Fell 1987).

The nature of this 5-year alcohol reduction trend was examined in the following
manner. Specific decreases of certain types of drivers and certain types of crashes were
compared with the overall reduction. If these specific reductions were substantially
greater than the overall reduction, then that would indicate that these drivers or
conditions were affected most. Figure 1 summarizes the key findings concerning the
nature of the reduction.

The largest reductions noted were for teenage drivers (28 percent), followed by

teenage pedestrians killed in collisions (26 percent). Also affected were drivers of vans
(23 percent reduction), female drivers (21 percent), and drivers who survived the fatal

Table 7. BACs for all drivers involved in fatal crashes, 1982-86

 

 

(in percents)
BAC 1982 1983 1984 1985 1986 1982-86 change
0.00 61 62 64 66 66 .
0.01-0.09 9 9 9 8- 8
0.10 and higher 30 29 27 26 26 -14

N- 56,029 54,656 57,512 57,883 60,297
BACKGROUND PAPERS

Table 8. BACS for teenage (16-19) drivers involved in fatal crashes

 

 

(in percents)
BAC 1982 1983 1984 19851986 1982-86 change
0.00 58 sé 63 67 66.
0.01-0.09. «13 13 3B on 8
0.10% and higher 29 27 24 -22 21 -28

N 7,467 7,050 7,366 7,151 7,854

 

crashes (17 percent). The absolute reduction was also larger in weekday crashes (17
percent) and in multivehicle crashes (16 percent).

Drivers age 25-34 had only a slight reduction during this 5-year period (6 percent).
Motorcycle drivers, with the highest percentage of alcohol involvement to begin with,
experienced no change in the percentage of drivers legally intoxicated during this period.
Pedestrians age 20-64 also had no reduction in the percentage legally intoxicated
between 1982 and 1986. Late night crashes and single-vehicle crashes showed only
modest reductions in the percentage of drivers who were at 0.10 BAC or higher (6
percent and 9 percent, respectively).

The average BAC of drinking drivers in fatal crashes in States where most of the

drivers were tested showed a modest decrease from 0.165 in 1980 to 0.153 in 1986.

Alcohol consumption per capita decreased in the United States between 1982 and
1986. But if that decrease was a prime factor in the decreased alcohol involvement of
drivers in fatal crashes, then a similar reduction should have occurred in intoxicated adult
pedestrians in fatal crashes, which was not the case.

 

 

 

 

 

 

 

 

 

 

 

 

0 10 20 30
All drivers ee 137%
Teen drivers re peered 28%
Van drivers ..  _On-7xzTVTLL_—__ I BM
Female drivers Cc — 1 21%
Surviving drivers Cc —J 21%
Teen pedestrians Cc __. —"] 26%
Daytime crashes (6 a.m.-6 p.m.) 17%
_ Weekday crashes 17%
Multivehicle crashes 16%

 

Figure 1. Nature of alcohol reduction in fatal crashes, 1982-86
Decrease in percentage drunk (BAC =0.10 or higher)
EPIDEMIOLOGY AND DATA MANAGEMENT

Table 9. Nature of alcohol reduction among fatally injured drivers with
_ known BACs in 15 good reporting States*

 

 

(percents) |
BAC of fatally 1980 1981 1982 1983 1984 1985 1986 1980-86
injured drivers . . . change |
00 39 40 a $B
01-.09 10 11 10 11 10 12 +9
10-.19 7 6 86 86 B BW RB ~~ -19
20+ 24 2B 23 21 20 19 £418 #8 ~25

 

*Test and report BACs on 85 percent of fatal drivers.

The nature of the reduction of alcohol in fatal crashes does seem to point to main
effects in responsible social drinkers, i.e., substantial reductions in daytime crashes, by
female drivers, drivers of vans, and teenagers. However, there is some evidence that the
percentage of drivers with very high BACs is also decreasing, at least in the 15 “good-
reporting” States in FARS. Table 9 shows that, in 1980, almost a quarter (24 percent) of
the fatally injured drivers in these States had BACs of 0.20 or greater. That portion in
1986 was 18 percent, which was a 25-percent reduction — greater than the reduction for
the drivers at BACs between 0.10 and 0.19. Most researchers would agree that drivers
at 0.20 BAC or greater are most likely problem drinkers or alcoholics. Yet the percentage
of drivers at these levels has decreased significantly since 1980 (in that 15-State sample).
Are these problem drinkers finding alternatives to driving? Are they confining their
drinking to their homes? Have many of them stopped drinking? More research is
necessary to answer these important questions.

Alcohol in Noncrash Drivers

Accurate determination of alcohol actually present in drivers while they are using the
highways can be estimated only by obtaining measurements from samples of these drivers
at roadside. (Thus, self-reported drinking-and-driving data from telephone or household
surveys are not considered here.) Measurement of alcohol in noncrash drivers is general-
ly obtained at roadside for four major purposes: oo

1. To estimate the contribution of alcohol to crash risk

2. To provide data for describing a particular problem by identifying and specify-
ing relevant parameters

3. To provide data for evaluating the results of any changes in circumstances
surrounding the particular problem, whether they result from unplanned
natural events or from controlled countermeasures

4. To foster general deterrence of drunk driving and to enforce DUI laws.

Research designed to accomplish the first purpose involves case-control studies.
Activities designed for the fourth purpose are currently referred to as either enforcement
checkpoints or sobriety checkpoints. Studies designed for the second or third purpose
have a broader range of objectives. Useful epidemiologic data can be obtained from
activities designed for any of these four purposes, but the most fundamental question is
addressed in investigations of alcohol and crash risk by means of case-control studies.
BACKGROUND PAPERS

Case-Control Roadside Surveys and Alcohol Crash Risk

That alcohol is found in approximately 50 percent of fatally injured drivers tested does
not necessarily prove that alcohol actually contributed to the occurrence of these crashes.
To begin building a case for or against the actual contribution of alcohol, it is first
necessary to determine the extent to which fatally injured drivers with alcohol are
representative of drivers with similar exposure, but not involved in the crashes. Thus, it
is necessary to compare the distribution of BACs obtained from control or comparison
drivers randomly selected while passing the same place as the crashes and at equivalent
times. By comparing these two sets of data, it is then possible to determine the similarities .
and differences between the two sets of drivers in terms of the percentages of each with
no alcohol, with detectable alcohol, with medium BACs, with high BACs, and so forth.

For example, a number of studies have indicated that between 40 and 50 percent of
fatally injured drivers examined had BACs of 0.10 or higher. If we had been able to
examine the other motorists who were actually driving at the same times and places that
these fatal crashes occurred, and if we had found that about 45 percent of these
noncrash-involved drivers also had BACs of 0.10 or higher, then we would have no basis
for concluding anything at all about the contribution of alcohol to highway crashes. That
is, the percentages of high-BAC drivers in the fatally injured sample would have been
the same as the percentage of high-BAC drivers in the comparison sample from the
population-at-risk — namely, about 45 percent. Therefore, in this hypothetical instance,
high-BAC drivers would have been neither under- nor overrepresented in terms of the

percentage of the population-at-risk made up of high-BAC drivers, namely, about 45
percent.

Conversely, if we had found a significant difference between the percentage of
high-BAC fatally injured drivers and the percentage of high-BAC control drivers from
the same population-at-risk, then we would be able to make some strong inferences about
the relative contribution of alcohol to these fatal crashes. This line of reasoning provides
the logical basis for attempting to obtain these BAC data from the population-at-risk
using the case-control design with roadside research surveys.

The first such study was conducted in Evanston, Illinois, 50 years ago by Holcomb
(1938), and several more studies have been conducted in the United States and abroad
since that time. These case-control studies have been analyzed from a variety of perspec-
tives and summarized in a number of publications (Hurst 1973, 1985; NHTSA 19835;
Perrine 1975a, b; Reed 1981; Zylman 1971). However, the material that follows in this
subsection is taken primarily from the most recent review (NRC 1987). In all these
reviews, a consistent pattern is revealed by the case-control studies: crash risk increases
sharply as BAC rises.

The relative probability of being involved in a crash is defined as the ratio of the BACs
of comparison drivers to those of drivers involved in crashes. This probability remains
roughly equivalent for crash-involved drivers compared with noncrash drivers up to
about 0.08 BAC (figure 2). (However, these relative risk curves understate the risk of
involvement at low BACs.) Although the rate of increased risk varies across studies (in
part because some studies examine all crashes and some examine only fatal crashes), the .
risk increases after about 0.08 BAC in all cases and increases dramatically after 0.10
BAC in most studies.

The curves depicted in figure 2 are based on groups of drivers of different ages who
have varying experience with alcohol and with driving. Because of the heterogeneity of
control groups and the lack of perfect comparability, the effect of alcohol at low BACs
is masked by other variables. For example, the major shortcoming of the Grand Rapids
study (among the most cited case-control studies) is the lack of comparability between
the drivers involved in crashes and the control drivers regarding the frequency of
consuming alcohol. This lack of comparability is the source of the apparent improvement
in crash risk at low BACs in the Grand Rapids data (the much debated “Grand Rapids
EPIDEMIOLOGY AND DATA MANAGEMENT

@—— Grand Rapids data (5,985 total crashes)

@——-8 Grand Rapids data (300 total or serlous crashes)
O———c Evanston data (270 injury crashes)

C-—-O Toronto data (423 total crashes)

@—— Manhattan data (34 fatal crashes)

30 eeseces Estimated approximate extension of Manhattan trace
Oo——O Vermont data (106 fatal crashes)

— a——4 Huntsville data (615 Injury crashes)

26 a——-—4 Adelaide data (299 Injury crashes)

RELATIVE PROBABILITY AT ZERO ALCOHOL

 

RELATIVE PROBABILITY OF CRASH INVOLVEMENT

WHERE 1.0=

 

 

0 02 .04 06 08 10 12 .14 16 .18 .20

BAC %W/V

Figure 2. Relative probability of crash involvement as a function of BAC. (Hurst
1985.) Reprinted with permission.

dip”) and the understatement of the risk of crash involvement at low BACs. Hurst (1973)
noted that the control group had a higher percentage of drivers who were regular
consumers of alcohol. Their apparently greater tolerance for alcohol had made them
safer drivers at low BACs than the drivers involved in crashes at low BACs, presumably
because the latter had less experience as drinkers. Hurst recalculated the relative risk
of crash involvement in the Grand Rapids data based on the drivers’ self-reported
frequency of alcohol consumption (figure 3). He drew three conclusions from the results.
First, drivers with frequent experience as drinkers are less likely to be involved in crashes
than light and medium drinkers at comparable BACs. Second, regardless of the
tolerance for alcohol, the risk of crash involvement increases with BAC. Third, the curves
greatly underestimate the risk for the average driver at any BAC; they only demonstrate
the relative hazard to drivers who regularly drink and drive. The curvilinear relationship
between relative risk of crash involvement and BAC is therefore caused in part by the
comparison of drivers with varying degrees of experience as drinkers and experience
driving under the influence of alcohol. When experience with alcohol is controlled for,
the risk of crash involvement increases with BAC without evidence of a threshold effect.

__ As noted by Perrine (1975p) in his review of the literature, the relative risk of
involvement is not the same as evidence of causality. Given the many interacting factors
that may contribute to a crash (and the lack of data on many of them), the role of any
single factor is difficult to isolate. Three of the case-control studies deserve special

attention because they also estimate the effect of alcohol on the probability of being
responsible for a crash.

The methodology for estimating crash responsibility was first developed by McCarroll
and Haddon (1962) in their case-control study of fatal crashes in Manhattan. They

45
46

BACKGROUND PAPERS

10.0 - SELF-REPORTED DRINKING FREQUENCY
B.0 | eeesees YEARLY OR LESS ra

| =e ~MONTHLY oe
6.0 ~——-—— WEEKLY 2 /

| mee OX/WEEK 2/7
40- ~~ DAILY ZS L

. ¢ -
= . / ”
a .

RELATIVE PROBABILITY OF INVOLVEMENT

 

 

 

BAC, %W/V

Figure 3. Relative probability of crash involvement by self-reported drinking fre-
quency (Hurst 1973). Reprinted with permission from the Journal of Safety Research,
a joint publication with the National Safety Council and Pergamon Press, Ltd.

categorized the crashes into five classes, the first three of which were assigned
responsibility:

1. Only one vehicle involved
2. Two vehicles involved but only one moving

3. More than one vehicle involved and in motion, with responsibility assigned
based on circumstances of the crash (cases in which there was any doubt were
excluded from this category).

The Manhattan study is based on a sample of 43 drivers fatally injured in crashes that
occurred between June 1950 and June 1960. For the 26 drivers in the assigned respon-
sibility classes, 19 (65 percent) had positive BACs, of which 14 (46 percent) had BACs
greater than 0.10 . Of these 14 drivers, 12 had BACs of 0.25 or greater. Of 156 drivers
randomly selectéd as controls at or near the sites of the crashes, 39 (25 percent) had
positive BACs, of which only 8 (5 percent) were at or above 0.10.

The Grand Rapids investigation involved by far the largest sample of all the case-con-
trol studies (5,985 crashes of all types) (Borkenstein et al. 1964, 1974). By comparison,
the 423 cases in the Toronto study constituted the next largest sample (Lucas et al, 1955;
Hurst 1985). Using McCarroll and Haddon’s method for assigning responsibility,

Borkenstein, Crowther, Shumate, Zill, and Zylman estimated that 3,305 of the involved |

_ drivers were responsible for the crash that occurred. They used the innocent drivers as

controls.

The Vermont study was the third to estimate crash responsibility, based on 106 cases
(all fatal crashes) (Perrine ct al. 1971). These crashes resulted in 113 fatalities, and 97 of

the drivers were assigned responsibility, again relying on the method developed by

McCarroll and Haddon. Of the drivers judged responsible, 60 percent had positive
BACs and 46 percent had BACs at or above 0.10. Perrine, Waller, and Harris (1971)
also calculated a crash-responsible curve, but in contrast to the Grand Rapids study, the
drivers stopped at roadblocks were used as controls.
EPIDEMIOLOGY AND DATA MANAGEMENT

m———4 ‘GRAND RAPIDS DATA, 3305 CRASHES,
DRIVER ASSUMED RESPONSIBLE

@——® MANHATTAN DATA, 24 FATAL CRASHES,
DRIVER ASSUMED RESPONSIBLE

O——O VERMONT DATA, 75 FATAL CRASHES,
DRIVER ASSUMED RESPONSIBLE

 

RELATIVE CRASH PROBABILITY

 

 

02 04 06 08 10 12 14 16 «18 20°
BAC % (W/V) OF DRIVER

Figure 4. Relative crash responsibility for drivers assumed responsible and those
not assumed responsible as a function of BAC, where 1.0 = relative probability at zero
alcohol. (Hurst 1973.) Reprinted with permission from the Journal of Safety Research,
a joint publication with the National Safety Council and Pergamon Press, Ltd.

Hurst (1973) replotted the curves from these three studies on a logarithmic scale to
facilitate comparison (figure 4). Although risk of crash responsibility increases as BAC

increases in all three studies, several disadvantages with the underlying data should be
noted.

The trend in the Manhattan data is based on a very small number of crashes: 25
responsible drivers with positive BACs. In addition, the trend at the higher BACs is
greatly understated. For the fatal crashes in which the driver had a BAC of 0.25 or higher
(about half those in the driver-responsible category), no driver in the control group had
an equivalent BAC. “Hence, the relative hazard calculated from the case/control ratio
would be infinite within the range, were it possible to graph it” (Hurst 1973).

One of the shortcomings of the relative risk curve estimated in the Grand Rapids
study is the inclusion of drivers involved in single-vehicle crashes in the group of
Tesponsible drivers. Although the responsibility of the driver is not in question, because
of the nature of the crash, a control driver is not available. The published data do not
Provide sufficient detail to allow the curve to be completely recalculated without the
single-vehicle crashes to determine the effect of including these crashes, but the available
data suggest that the curve would shift to the right. It would still accelerate after 0.04

feu and at an exponential rate, but the curve would not rise as quickly as shown in
re 4,

47
48

BACKGROUND PAPERS

One problem with the Vermont data is the small number of crashes in the sample. In
the comparison of crash risk as BAC rises, one. or two drivers are responsible in some
of the BAC ranges. Chance occurrence could distort the results when_so few drivers are

‘the basis of the calculations.

Despite the weaknesses in the case-control studies, some important conclusions can
be drawn. In several of the case-control studies done in the United States and abroad, a
consistent increase in risk of crash involvement has been shown. When experience with
drinking is controlled for, this risk increases with BAC without any evidence of a
threshold effect (or dip). The three studies that attempted to estimate crash respon-
sibility showed that the risk of causing a crash increases even more rapidly than the risk
of crash involvement as BAC increases (NRC 1987).

Another important aspect of the alcohol contribution to crash riskis reported by Voas
(NHTSA 1985). To emphasize the significance of the difference in BAC between drivers
assumed to be responsible versus those assumed not to be responsible for crashes, Hurst
(1974) also presented an additional calculation on the data from the Grand Rapids study.
His results for the drivers assumed to be responsible are represented by the center plot
in figure 4. However, the probability of being innocently involved in a crash remains
essentially level and does not increase with increasing BAC; the plot is basically flat and
would lie between the relative crash probability of 1 and 2 in figure 4. The same result
was found in the Huntsville/San Diego study by Farris, Malone, and Lilliefors (1977).
These results provide further evidence for the causal role of alcohol in crashes.

Other Roadside Research

The success and utility of the case-control procedures for investigating alcohol crash
risk stimulated interest in using the roadside survey technique for evaluating alcohol
safety programs by measuring the change in the number of high-BAC drivers actually
on the roads. Standardized procedures for conducting roadside surveys were developed
(Perrine 1971) and applied successfully in 28 of the 35 Alcohol Safety Action Projects
(ASAPs) funded by the Department of Transportation between 1970 and 1975 (Voas
1972; Carr et al. 1974). In these programs, the roadside surveys were used to evaluate
project effectiveness (Levy et al. 1978) by serving as a means to collect data used as the
primary criterion or dependent variable (BAC). Roadside surveys were conducted
before and after program implementation to measure the change in average driver BAC
(if any) resulting from project activities (Lehman et al. 1975). When used for program
evaluation, sampling was conducted during periods when a high percentage of drinking
drivers was on the road (i.e., Friday and Saturday nights) rather than at times and places
at which accidents had occurred. By a return to the same sites, changes over time can be
measured. Se

Roadside surveys provide a more direct method of evaluating alcohol safety counter-
measure programs than does the use of accident data, because highway crashes result
from a large number of factors (weather, roadway construction, economic conditions,
etc.) that are unrelated to the evaluation of enforcement activities. The BAC values of
drivers serve as an intermediate measure between action programs and the ultimate
criterion of accident prevention. While a reduction in the average BAC of drivers on the
road does not guarantee a reduction in crashes, the relationship between driver BAC

- and risk of crash involvement is close enough to make this measure a credible criterion

for program effectiveness.

During the ASAP period (1970 through 1974), some 77 roadside breath-testing
surveys of nighttime drivers were conducted. In addition, a national roadside breath-
testing survey was conducted in 1973, and a computer archive of these 78 roadside
surveys is stored at the University of Michigan Highway Safety Research Institute
(Lehman et al. 1975). The file contains breath-testing results, demographic data, and so
forth, for some 78,000 randomly selected drivers, as well as 2,700 passengers. Analysis
EPIDEMIOLOGY AND DATA MANAGEMENT

of these aggregated data show the following percentages of drivers with BACs at or
exceeding 0.10: 1 percent of weekday-early drivers, 3 percent of weekend-early drivers,
and 6 percent of weekend-late and. weekday-late drivers..Significant reductions in the
percentages of drivers above the legal limit (0.10 BAC) were demonstrated for those
jurisdictions that used this evaluation method (Levy et al. 1978).

Based on the utility of this roadside BAC measure in the ASAP program, it was
applied again in a 4-year study of a special DUI enforcement effort in Stockton,
California (Voas and Hause 1987). In this study, survey procedures were modified to
permit low cost and low profile surveys that were conducted every weekend for 3 ’” years
(Hause et al. 1982). Drivers with a BAC of 0.10 or greater on Friday and Saturday nights
decreased from 88 per thousand before the Stockton project to 50 per thousand during
the third year.

The roadside survey technique also permits (through an application of Bayes’
Theorem) estimation of the probability that a driver at a given BAC will be arrested by
the police. This procedure was first applied by Beitel, Sharp, and Glauz (1975) to the
ASAP survey data in Kansas City. Hause, Voas, and Chavez (1982) used the same
procedure in Stockton. These studies provided roughly similar results indicating that the
chances of being arrested at a BAC of 0.15 is roughly 1 in 100, while the chance of arrest
at 0.10 BAC is half that amount, about 1 in 200. Since both these studies involved intensive
enforcement programs, they provide a reasonable indication of the maximum arrest rate
that can be achieved with traditional patrol methods.

The ASAP experience with roadside research surveys and with the success of the
manual for conducting and evaluating them (Perrine 1971) provided the basis for
subsequent international activity. An invitational international workshop was conducted
in Paris in an attempt to coordinate the methodology for roadside research surveys to
be implemented in other countries in order to maximize the comparability of the
obtained data. The workshop resulted in a useful manual (Carr et al. 1974) and comple-
mented parallel activities being conducted under the auspices of the Organization of
Economic and Cooperative Development. As a result of these activities, use of roadside
surveys for international comparisons of countermeasure programs was stimulated in
Canada, the Netherlands, Norway, Sweden, and Finland. The resulting data permitted
an international comparison of driver BACs (Voas 1982), which indicated that ap-
proximately 12 percent of drivers on weekend nights were at or above 0.05 BAC in
Canada, the Netherlands, and the United States, whereas less than 2 percent of drivers
were at this level in Scandinavian countries.

Although use of roadside research surveys has diminished in the United States since
the end of the ASAP activities in the mid-1970s, the technique continues to be used
effectively in other nations, for example, Canada, Australia, Denmark, and Norway.
Nevertheless, a few studies: of the population-at-risk in the United States either have
been conducted recently or are currently being conducted. In the spring of 1986, U.S.
National Roadside Breathtesting Survey II (Wolfe 1986) was conducted in a repre-
sentative sample of 32 localities, 18 of which had participated in the 1973 U.S. National
Roadside Breathtesting Survey I (Wolfe 1974). Statistically significant reductions were
found in the percentage of medium and high BAC drivers sampled at high-risk times
(Friday and Saturday nights from 10 p.m. to 3.a.m.). Drivers at or above the illegal BAC
of 0.10 decreased from 5.0 percent in 1973 to 3.1 percent in 1986; drivers at or above a
BAC of 0.05 decreased from 13.5 percent in 1973 to 8.3 percent in 1986. It should be

noted that the breathtest completion rates were 86 percent in 1973 and 92 percent in
1986. ~

In Vermont, a large-scale roadside research study involving a projected 42,000

_ nocturnal drivers sampled at high-risk times (Friday and Saturday nights from 10 p.m.
to3.a.m.) is currently being conducted. It is funded by the National Institute on Alcohol
Abuse and Alcoholism (NIAAA, Grant AA07876). This 5-year field study is primarily

49
50

BACKGROUND PAPERS

designed to determine the prevalence of drivers with high alcohol tolerance, and to
determine their salient and differentiating characteristics. Since the high alcohol tolerant
driver is apparently rare, a large number of motorists (42,000) driving at high-risk times
must be stopped and screened for BAC to identify a sufficient number of such people
(40 to 60) to be able to conduct a meaningful study. In the process of conducting this
field study, a large number of people will be bréathtested using both the new passive
alcohol sensor and the more traditional hand-held evidentiary devices. Approximately
4,000 of these motorists, sampled across the full distribution of BACs, will participate in
extensive. personal interviews concerning self-reported background data; drinking, driv-
ing, drinking-and-driving, drugs-and-driving information and attitudes; and selected
personality characteristics. Data will also be gathered on these motorists’ driver records,
performance on the most valid field sobriety tests (gaze nystagmus, walk-and-turn, and
standing steadiness), and ratings on clinical signs of intoxication. With a test completion
rate of 96 percent, the results from the first 650 drivers indicate that 3.7 percent had a
BAC of 0.10 or higher, whereas 10.2 percent had a BAC of 0.05 or higher. Although the
sample size is.still relatively small, these 1988 data show a decrease in distribution of
BAC when compared with data obtained in a 1974 Vermont study (Perrine 1976) of 1,663 :
drivers at high-risk nocturnal times (Thursday, Friday, and Saturday nights between
10:30 p.m. and 3 a.m.): 4.6 percent had a BAC of 0.10 or higher and 14.7 percent had a
BAC of 0.05 or higher. Thus, these roadside studies of the high-risk population would
seem to show that some progress is being made in the war on drunk driving, if motorists
with BACs in excess of the legal standard (0.10) are taken as the criterion.

Enforcement Checkpoints

Police officers conduct sobriety checkpoints at which they stop motorists at random
and test them for breath alcohol. Such activities are conducted primarily for enforcement
purposes, although they also serve as general deterrence. Although useful data for
epidemiologic purposes are available from these enforcement checkpoints, few sys-
tematic studies have been conducted to analyze such data. Ifthey were analyzed carefully
and properly, these data could provide a valuable source of relatively low-cost informa-
tion concerning the population-at-risk. In a recent Charlottesville, Virginia study, Voas,
Rhodenizer, and Lynn (1985) evaluated the effectiveness of such sobriety checkpoints,
especially in comparison with the drivers arrested for DUI by traditional roving patrols.

- In addition, this study found evidence of police biases in those arrested, whereby both

young drivers and women were underrepresented among arrested drivers, while minority
and very high BAC drivers were overrepresented. Thus, such studies clearly demonstrate
that researchers can avail themselves of enforcement checkpoints as an opportunity to
collect valuable data for epidemiologic purposes.

Characteristics of Drunk Drivers

During the past 20 years, numerous statistical and clinical studies have been published
on various aspects of the drinking driving problem. However, surprisingly little rigorous
research has been published on characteristics of convicted DUI offenders, particularly
when contrasted with the vast literature on problem drinkers/alcoholics, and on charac- -
teristics of drivers involved in fatal accidents. As suggested by Zylman (1974) and by
Moskowitz, Walker, and Gomberg (1979), the population arrested and convicted of DUI
offenses is not typical of impaired drivers in general or of drivers involved in alcohol-re-
lated accidents. The mean BAC of convicted DUI offenders in California during 1984
was 0.18—a concentration far in excess of the State’s 0.10 limit per se, and well beyond
the level at which impairment occurs. In one of the few formal statistical studies of
differences between alcohol-involved fatal accident drivers and convicted DUI of-
fenders, Fridlund and Hagen (1977) used discriminant function analysis in comparing
146 DUI offenders in Los Angeles County with a sample of 191 alcohol fatalities. The
DUI conviction group had significantly more prior DUIs, more prior reckless-driving
EPIDEMIOLOGY AND DATA MANAGEMENT

convictions, and more prior moving traffic convictions. The differences on the incidence
of DUIs and reckless convictions was large, with the DUI group having about three times
as many entries during the prior 3-year period. |

Moskowitz, Walker, and Gomberg (1979) conducted a detailed review of the litera-

ture on DUI offender characteristics. This subsection relies heavily on their monograph
for the pre-1979 literature, but concentrates on recent studies and studies not included
in the 1979 review for its primary source references. However, a few pre-1979 studies of

special importance are reviewed here as primary references even though they are also
included in Moskowitz et al. (1979).

Moskowitz, Walker, and Gomberg organized their review by type of offender char-
acteristic (prior driver record, age, etc.), and reached the following conclusions with

respect to each domain:

e Marital status: DUI convictees are much more likely to be divorced, separated,
or widowed than are non-DUI control populations. Some studies have reported
five- to sixfold differences in rates compared with control populations.

e Employment history: Convicted DUI offenders are more likely to be un-

employed, with rate differentials ranging from two- to fourfold higher across
various studies.

e Occupation: Convicted DUI offenders are more likely to have lower status
-occupations. DUI offenders in blue collar jobs averaged 65 percent across
studies compared with 51 percent for control samples.

e Income: Convicted DUI offenders tend to have lower incomes—about 18
percent lower than controls across the reviewed studies.

e BAC: The mean BACs for the offenders averaged from 0.18 to 0.28. (Statewide
California figures have consistently averaged 0.18.)

e Drinking behavior: DUI offenders drink more often and consume more alcohol
per sitting than do non-DUI populations. Beer is the preferred beverage of DUI
offenders. The findings of the Southern California study by Pollack (1969) are
typical. Pollack reported that 18 percent of DUI drivers drank every day
compared with 11.5 percent for a control sample. In terms of drinks per sitting,
35.2 percent of the DUI sample typically consumed five or more drinks com-
pared with 5.7 percent of the control sample.

e Reason for drinking: Convicted DUIs (and alcoholics) are more likely to drink
to release tension and to cope with stress. ,

e Problems caused by drinking: Convicted DUIs are much more likely than
controls to exhibit poor health, family disorganization, financial problems, and
poor job performance.

e Prior alcohol treatment history: Convicted DUIs are more likely than controls’

to have previously entered some form of alcohol treatment program. The
median across 20 studies was 6.0 percent, with a maximum of 42.5 percent. (This
characteristic is highly dependent on the institution and delivery systems of a

particular region, and would be expected to vary across jurisdictions and over
time.)

e Problem drinking status: Studies using the Michigan Alcoholism Screening Test
(MAST) indicate that 54-74 percent of convicted DUIs fall in the problem-
drinking and alcoholic range. Studies using the Mortimer-Filkins test produce
slightly lower prevalence figures.

e Driving after drinking: Convicted DUIs are much more likely than controls to
drive after drinking. Pollack (1969), for example, reported that 49 percent of
DUI offenders admitted to driving at least once a week after two drinks
compared with 12 percent of a control sample.

51
52

BACKGROUND PAPERS

e Total prior. arrests: Convicted DUI offenders are more likely to. have prior
arrests for both alcohol and nonalcohol offenses. :

e Driving history: Convicted DUI offenders have substantially more driving
record entries of all types than do controls (more DUIs, more alcohol and total
accidents, more moving traffic violations, and more license actions). The rate —
increases across studies and variables range from 100 percent to 500 percent.
The driving record histories of convictéd DUIs are also substantially worse
those of medically diagnosed alcoholics.

e Personality traits: Convicted DUIs have a significantly higher prevalence of
personality trait disorders. They are more likely than controls to. exhibit
neuroticism, depression, paranoid ideation, low self-esteem, and to have alower
sense of personal responsibility and control and greater feelings of aggres-
sion/hostility. ; ,

e Stress: Convicted DUI offenders are more likely to report experiencing stress
from family, financial, and job problems. *

e Education: Convicted DUI offenders are more likely to be high school dropouts
and have fewer years of education. .

e Age: Convicted DUI offenders tend to be slightly older than non-DUI controls,
with the highest disproportionate concentration in age interval 30-45.

e Race: Most convicted DUI offenders are white, but minority groups (hispanics ~
_ and blacks) are overrepresented compared with their representation in the
population.

e Sex(not covered by Moskowitz): The great majority of convicted DUI offenders
are male, The range for females is from 5 to 20 percent, depending on the
characteristics of the specific DUI population (first versus repeat offenders),
region, and so forth. A large statewide sampling in California indicates that 13
percent of those convicted for a DUI offense in 1982 were female (Tashima and
Peck 1986). Among first and second offenders, females accounted for 17
percent and 10 percent, respectively.

Drinking Status

It is clear from the above summary that persons convicted of drunk driving offenses
deviate greatly from the general driving population on a wide variety of characteristics.
That DUI offenders contain a disproportionate number of problem drinkers would be
expected, of course, since the offense of drunk driving is, per se, a problem associated
with the consumption of alcohol.

The number of drinks required to produce manifestly detectable impairment in
driving and the BACs typically attained by DUI offenders implies a level of alcohol
consumption that is statistically deviant. Given the very low probability that any given
incident of impaired driving will result in detection (arrest or accident), the percentage
of DUI offenders who were simply unlucky, in the sense of getting caught in a rare
instance of impaired driving, would be relatively small. One would therefore expect most
DUI offenders to be heavy consumers of alcohol.

Most of the empirical literature, and most authorities in the area, agree with this
conclusion, although controversy has developed over the percentage of DUI offenders
who are alcoholics in the clinical disease context. This controversy stems more from
semantic and epistemological complexities than from disagreements over data, and is
not pursued here.

Lest the impression be created that opinion and data are unanimous on the drinking
status of DUI offenders, the results of a recent California study of first offenders will be
summarized in detail. This study was carried out by the Pacific Institute for Research
EPIDEMIOLOGY AND DATA MANAGEMENT

and Evaluation (PIRE); it was commissioned by the State of California pursuant to
Assembly Bill 3405 (Stewart et al. 1987).

The major objective of the PIRE study was to develop a model curriculum and _

rehabilitation program for use by courts in sentencing first-offender DUI cases. How-
ever, this review will only consider that component of the study pertaining to offender
characteristics (natural variation component). ,

Detailed biographical, drinking habit, and arrest-incident information was collected
by questionnaire on 5,052 respondents from 26 first-offender treatment programs
throughout California. The authors reported the following statistics from an analysis of
the questionnaire responses: |

e Median number of drinks on day of arrest: 6

e Median BAC upon arrest: 0.16 ;

e Median number of days in past year with four or more drinks: 60

e Median number of days in past year with eight or more drinks: 4

e Percentage who did not feel intoxicated upon arrest: 35 percent

» Median number of previous days in past year driven while impaired: 1
« Percentage with prior DUI arrests: 20 percent

The authors categorized the drinking pattern responses into two typologies for
comparison with a statewide general population survey. The more complex of the
typologies was a 7-point continuum: abstainers, infrequent drinkers, light monthly, light
weekly, moderate weekly, and frequent heavy. Using a probit analysis to adjust for
population differences, the authors found no significant differences between the drink-
ing frequency of the two groups after abstainers were removed from the general
population sample.

The difference was significant, however, in the frequency of heavy drinking (five or
more drinks at least once a week), with substantially more of the DUI offenders falling
into that category. Nevertheless, fewer than 30 percent of the first offenders were placed
in this category. In commenting on these findings, the authors concluded:

While defining a “typical pattern of drinking for all subjects is difficult, the
median frequencies of use conditional upon the level of use at arrest is
revealing: 40 percent of the subjects reported having had 8 or more drinks
on the day of their arrest. The median frequency of use at this level among
these subjects was 14 occasions over the previous year and only one occasion
in the preceding 30 days. A pattern of use including drinking 8 or more drinks
at least weekly over the previous year was reported by only 20 percent of
these subjects. Thus, for the majority of subjects, drinking at the level of use
at which they were arrested is relatively infrequent (pp. 27-28)

.- . first offenders are not unlike the general population of drinkers in
California in terms of the typical frequency of use. However, it appears that
the incidence of heavier drinking is greater among first offenders (p. 32)

The authors also included a measure of alcohol dependency in their study. Subjects
completed a 25-item Alcohol Dependency Scale (Skinner and Allen 1982), and the
scores were compared with those of clinically diagnosed alcoholics. Ninty-two percent
° re subjects produced scores “indicating a low level of alcohol dependency.” The

uthors went on to conclude “the dramatic differences in these distributions suggest that

depe :
dependency symptoms among first offenders are quite low, as compared to alcohol
treatment groups.” .

and results and conclusions of the PIRE study are at odds with prevailing opinion
ost prior studies in this area. If the findings are accepted at face value, the great

53
54

‘ BACKGROUND PAPERS

majority of first offenders fall within the bounds of social drinking. Even the percentage
characterized as heavy does not seem extreme. , o

There are several possible explanations. First, the PIRE study was limited to first
offenders, and it is known that the percentage of problem drinkers among first offenders

is lower than among repeat offenders.

Second, problem drinkers and DUI offenders are known to understate their drinking,
sometimes dramatically. Stewart, Epstein, Greenewald, Laurence, and Roth (1987)
acknowledge this possibility and recommend caution in interpreting the study findings.
Although previous studies are also subject to reporting biases, many have included
clinician interviews, psychometric instruments employing lie scales, and a variety of
public agency data. No evidence is presented in the PIRE study to indicate that
procedural controls were used to minimize the tendency of people to “fake good” or
employ various forms of self-denial.

Third, the first-offender survey only involved offenders who were sentenced to an ©
alcohol program and who agreed to cooperate by completing the questionnaire. In
California, 23 percent of first offenders are not assigned to programs. Approximately 15
percent of the sample did not return a questionnaire.

There are also inconsistencies in some of the values derived from the self-report. For
example, it would have taken more than a median of six drinks to produce a median BAC

0.16. The fact that 35 percent of the subjects did not feel intoxicated when arrested and

that only 14 percent acknowledged being definitely intoxicated is cause for further
suspicion.

Finally, it is difficult to accept the median estimate of only one incident of driving
while impaired in the previous 12 months. The suggestion is that many of the subjects
were not being candid in their responses.

The PIRE report also contains a description of first offender biographical and

socioeconomic characteristics. The statistics of interest are summarized below:

e Male: 81 percent

e Single: 46 percent

e Divorced, widowed, or separated: 21 percent

e White: 68 percent

e Hispanic: 22 percent

e Black: 3 percent
e High school dropouts: 22 percent
e Median age: 30 -
e Unemployed or employed part-time: 30 percent
e Median income: $16,500

These demographic characteristics are reasonably consistent with the portrayal from
the Moskowitz review, particularly when allowance is made for differences in time and

“region. The percentages for ethnic minorities are somewhat lower than would be

expected based on California ethnicity composition and prior evidence showing that
some minorities (e.g., Hispanics) are overrepresented in DUI populations. It is impor-
tant to recognize that the PIRE sample is limited to offenders entering first-offender
programs and, within this subset, to those who returned the questionnaire. These factors
could alter the representativeness of the sample.

Prior Driving Record

It is also clear that DUI offenders have statistically deviant driver records before their
EPIDEMIOLOGY AND DATA MANAGEMENT

DUI arrest. Although one would expect overinvolvement in previous alcohol-related
accidents and convictions, the extent of DUI offender overinvolvement in nonalcohol
related incidents is not widely recognized.

Statewide, data from Tashima and Peck (1986) indicate the following pretreatment

30-month rates for representative statewide samples of 29,097 first and 7,797 repeat DUI
offenders:

Mean non-DUI Meantotal Mean non-DUI related

 

accidents . accidents convictions
First offenders . 17 36 13
Second offenders 18 rc) 1.4

The above rates are more than twice the rates expected for asimilarly stratified (age and
sex) population of non-DUI drivers.

The Tashima and Peck (1986) study and numerous previous California studies
indicate that nonsuspended DUI offenders also accumulate worse non-DUI driving
records (total accidents, total moving violations, etc.), following conviction for a first or

repeat DUI offense (Sadler and Perrine 1984; Hagen et al. 1978; Hagen 1977; Arstein-
Kerslake and Peck 1985).

The most detailed analysis was performed by Arstein-Kerslake and Peck, who
compared the subsequent 4-year driving records of first and repeat DUI offenders from
Sacramento County with a general population group that was similarly age-sex stratified.
The first and repeat offenders were grouped into quartiles based on their actual and
predicted DUI recidivism. Except for the first quartiles (i.c., lowest 25 percent in terms
of recidivism expectancy), all quartiles had substantially worse accident and traffic

conviction records. The differences were also highly significant when summed across
quartiles.

The above relationship between DUI offenses and driving behavior in general has
been addressed by a number of other investigators (Maisto et al. 1979, Raymond 1971,
Denberg 1974). Donelson, Beirness, and Mayhew (1985) consider the issue from the
impaired problem-driver paradigm addressed in Simpson’s (1977) paper. This heuristic
paradigm views the convicted DUI population as containing drivers whose drinking is
subordinate to a larger problem of high-risk negligent driving. The alcohol impairment
can combine, additively or synergistically, with negligent driving to increase risk, but the
underlying problem-driving behavior exists independent of alcohol.

Although Donelson, Beirness, and Mayhew stress the hypothetical nature of this
paradigm, the premise that impaired drivers who drive aggressively and unlawfully are
more likely to be apprehended is in no way hypothetical. It has also been established that
DUI offenders with a prior history of moving violations represent substantially greater
accident risks than DUI offenders with clean records (Sadler and Perrine 1984;
McConnell and Hagen 1980; Peck and Kuan 1983).

The linkage between problem driving and DUI offenses is the very essence of arecent
study by Donovan, Umlauf, and Salzberg (in press). These investigators followed the
driving records of 254 non-DUI-involved problem drivers over a 3-year period sub-
sequent to initial identification. Approximately 11 percent of the sample had a DUI
conviction during that period—a rate five times greater than that of the general male
driving population in Washington State. The study was replicated on a sample of 38,695
driver record files. The authors found that drivers with four or more moving violations
were greatly overinvolved in subsequent DUI offenses, with 16.9 percent receiving an
initial DUI conviction during a 3-year followup period. The DUI rate was particularly
pronounced for males under age 30.

55
56

BACKGROUND PAPERS

DUI Recidivism

How do first offenders compare to repeat offenders on the various characteristics
described above? This question is, of course, related to the question of recidivism
correlates and is best addressed by longitudinal recidivism studies.

The actual rate of recidivism cannot be determined in any general sense because it is
inextricably tied to the length of the followup period, the length of record retention in a
given State, the DUI arrest rate of a particular region or State, regional plea reduction
practices, and the effectiveness of DUI countermeasures. In California, approximately
35 percent of all DUI convictions each year involve drivers with prior DUIs within the
preceding 5 years.

Recidivism prediction was directly addressed by Ellingstad (1974) as part of the
evaluation of the South Dakota ASAP. Discriminant analyses, performed separately for
problem and nonproblem drinkers, assessed the predictability of a dichotomous DUI
recidivism measure using 14 variables related to prior conviction history, demographic
characteristics, drinking pattern, and Mortimer-Filkins score. The problem-drinker
group yielded the highest level of prediction. For the 1,744 problem-drinker clients, of
whom only 12.6 percent were actual recidivists, prediction of subsequent 2-year DUI
recidivism was significant at the 0.001 level. However, only 4.4 percent of the variance in
recidivism was accounted for by the discriminant function (multiple R =0.209). Of the
14 variables used for the recidivism analysis, only 6 had a significant univariate relation-
ship with recidivism: prior DUI convictions, reckless convictions, total convictions,
marital status, drinking pattern, and Mortimer-Filkins score. All relationships were in
the expected direction —that is, less favorable values were associated with increased
recidivism.

The level of recidivism prediction reported by Ellingstad (1974) was greater than that
reported by Burch (1974) in her analysis of the Los Angeles ASAP. A multiple regression
analysis was conducted for approximately 1,000 clients with the objective of predicting
subsequent 7-month DUI recidivism using treatment, age, accident, and conviction
measures as predictor variables. The actual 7-month recidivism rate in the sample was
approximately 10 percent. The analysis indicated that 2.4 percent of the variance in
recidivism could be accounted for by the seven predictor variables (R =0.155, p< 0.01).

The relatively low level of prediction reported by Burch results at least in part from the

brief 7-month period during which recidivism data were collected.

As part of the evaluation of the El Cajon Drinking Driver Countermeasure Program,
Wendling and Kolodij (1977) collected data on driving history, criminal arrest record,
probation officers’ evaluation of problem-drinking severity, and Mortimer-Filkins diag-
nostic scores from 1,740 DUI offenders. These measures served as predictors of the
yearly rate of recidivism (DUI and reckless-driving convictions). The duration of data
collection subsequent to: treatment ranged from 0 to 72 months. Stepwise multiple
regressions were performed for each half of the sample, and then each equation was
applied to the other half of the sample in order to provide a measure of cross-validation.
Although Wendling and Kolodij report impressive Rs in the range of 0.40, the high levels
of classification error upon cross-validation indicate that the construct multiple Rs were
inflated.

Development of a prediction model to identify likely recidivists among a sample of
Los Angeles DUI offenders was one of the main objectives of Pollack, Didenko,
McEachern, and Berger (1972). Three models were developed and evaluated: multiple
regression, discriminant function, and empirical bayes. The authors reported a high
degree of classification accuracy for drivers with extremely high predicted recidivism
expectancies. However, since these drivers represented only a small part of the total
recidivist population, it could not be concluded that recidivism can be accurately
predicted. On the contrary, the data indicated that the classification error would be
substantial for drunk drivers with nonextreme recidivism expectancies. Although no
EPIDEMIOLOGY AND DATA MANAGEMENT

multiple R or overall classification accuracy was cited, the authors reported that the best
prediction would achieve an 11-percent increase in predictive accuracy over what would
be expected by chance prediction. Increased recidivism was associated with lower
education, younger age, and a higher incidence of traffic accidents, traffic violations, and
nontraffic arrests.

McGuire (1975) assessed the predictability of accidents and alcohol-related convic-
tions for DUI 2,255 offenders who had participated in the Orange County Alcohol
Traffic Safety project. Convicted DUI offenders were assigned to one of a number of
alternative countermeasure programs. Driver record recidivism data were collected for
15 months after treatment assignment. McGuire performed stepwise regression analyses
of driver record, psychosocial, and psychometric variables to identify covariates sig-
nificantly related to accidents and alcohol-related convictions. The significant predictors
of subsequent accidents were sex, age, court-martial (if in service), number of accidents
in last 3 years, and number of traffic tickets in last 3 years, yielding a multiple R of 0.20.
Surprisingly, subsequent alcohol-related convictions were slightly less predictable than
subsequent accidents (R = 0.14). The significant predictors of alcohol-related convic-
tions were marital status, number of full-time jobs in last 5 years, frequency of smoking,
number of tattoos, and number of traffic tickets in last 3 years. No cross-validation
analysis was performed.

Arstein-Kerslake and Peck (1985) used multiple regression and discriminant function
techniques to predict the 4-year DUI rate, subsequent to treatment, of large samples of
first- and second-time DUI offenders selected from Sacramento County. The best
multiple R was 0.27, which shrank to 0.21 on cross-validation. The regression equations
and correlation coefficients indicated that recidivists were more likely to:

e Be younger

e Besingle or divorced

e Have more prior DUI offenses

e Have more nonalcohol moving traffic violations

e Bemale

e Have blue collar occupations

e Have more nonmoving traffic violations

e Be ethnic minorities

e Have previous alcohol treatment or disulfiram use

e Have exhibited negative attitude ratings during the intake interview

e Have received intake recommendations for more intensive alcohol
treatment

e Have higher BAC levels

Most of the above recidivism correlates are intuitively plauible and onsistent with the
prior literature. Simply put, DUI offeders are more likely to recidivate if their drinking
probem is more severe and their driving record reflects numeous non-DUI- and DUI-
related violations.

Arstein-Kerslake and Peck (1985) also developed regression models for predicting
program compliance — that is, successfully completing the rehabilitation program. Pro-
_gram compliance proved much more predictable than DUI recidivism. In addition,
persons with a high likelihood of being noncompliant tended to have extremely high
subsequent accident rates.

Multivariate and Taxonomic Studies of DUI Offender Characteristics

One limitation of simple univariate studies is that they fail to consider the inter-

57
58

BACKGROUND PAPERS

relationship among the set of variables being evaluated to characterize a given sample
of DUI offenders. Although regression and discriminant function procedures partial out
intercorrelations in producing linear composites that maximally differentiate between
groups (e.g., offenders versus nonoffenders), these techniques are focused on the sole
objective of discrimination with respect to a single statistical criterion. As a result, they
do not provide a portrait of the offender population in terms of the complete array of
measurements and the more general dimensions underlying those measures.

In recent years, a number of investigators have attempted to develop multivariate
typologies of DUI offenders through factor and cluster analysis procedures (Arstein-
Kerslake and Peck 1985; Wells-Parker et al. 1985). These efforts to construct empirically
anchored multivariate typologies were preceded by a number of attempts to produce
rational typologies through less formalized statistical or clinical methods. A brief sum-

- mary of this literature follows.

Intuitive and univariate typologies. The concept of distinguishing drinkers on a con-
tinuum of severity (social, problem, alcoholic; light, modest, heavy, primary versus
secondary alcoholism) has a long history. (Some research on DUI typologies has been

referred to above in connection with the literature on univariate characteristics of DUI
offenders.)

Cahalan, Cisin, and Crossley (1969) and J ellinek (1960) provide detailed examples of

‘analytic systems, based on a complex of medical, sociopsychological, and drinking-style

parameters. Many of the ASAPs developed clinical and statistical taxonomies for
classifying convicted drunk drivers. These efforts have been reviewed by Epperson,
Harano, and Peck (1975), Ellingstad (1974), and Nichols (1974). Although it is difficult
to formulate a coherent generalization about the success of these efforts because of the
diversity and limitations of the validation methods, the various systems had some utility
in elucidating characteristics of convicted DUI offenders and in differentiating offenders
from drivers in general. The following characteristics were often used to create
problem-drinking continuums, and each has been found to differentiate convicted DUI
offenders from non-DUI populations:

e Scores on psychometric and personality tests, such as the Minnesota
Multiphasic Personality Inventory (MMPI)

e Scores on tests specifically designed to detect problem drinkers and
alcoholism, such as the Mortimer-Filkins and MAST

e Quantity-frequency index scores

e Blood alcohol levels at time of arrest or accident involvement
e Prior record of DUI offenses and alcohol-involved accidents

e Prior arrests for public drunkenness and alcohol-associated
misdemeanors

e Other criminal offenses

Many of these variables were discussed above in connection with the univariate studies.

Nichols and Reis (1974) concluded that number of prior DUI offenses and BAC were
among the two most useful indicators for classifying DUI offenders into a problem
no-problem dichotomy, and they used this dichotomous system to classify DUI offenders
across many of the ASAP sites. A description of the complete classification criteria is
presented in table 10.

After 24 months of followup from the point of treatment classification, Nichols and
Reiss reported that 15 percent of the problem-drinker group had been rearrested
compared to 8 percent of the nonproblem group.

Epperson, Harano, and Peck (1975) regressed the BAC values for a sample of 1,366
EPIDEMIOLOGY AND DATA MANAGEMENT

Table 10. Department of Transportation problem drinker classification
criterion

1, Diagnosis as an alcoholic by a competent medical or treatment authority.

OR
2. Self admission of alcoholism or problem drinking.

OR
3, Two or more of the following: .
a. ABAC of 0.15 or more at time of arrest
b. Arecord of one or more prior alcohol-related arrests
c. Arecord of previous alcohol-related contacts with medical, social, or
community agencies -
d. Reports of marital, employment, or social problems related to alcohol
e. Diagnosis of problem drinker on the basis of approved structured written
diagnostic interview instruments (e.g., MAST, Mortimer-Filkins, National
Council on Alcoholism (NCA), Johns Hopkins diagnostic tests)

DULarrestees against a pool of several driver record measures obtained on each subject.
Higher BAC values were found to be associated with significantly higher (p< 0.05) rates
of prior alcohol-related accidents and DUI convictions. The authors concluded that a
combined criteria of prior offense frequency and BAC should be included in any
problem drinker driver taxonomy. In another component of the same study, they
reported that two psychometric tests, the Risk Addiction Profile and Mortimer-Filkins,
produced significant discrimination between a group of DUI arrestees with BACs in
excess of 0.20 and a group of non-alcohol-involved negligent drivers.

A number of authors have commented on the criterion problem in validating
problem-drinker taxonomies —a problem that emanates from the difficulties in defining
what constitutes problem drinking and alcoholism (Epperson et al. 1975). As a result,
different classification schemes and diagnostic procedures can diverge greatly in their
respective problem-drinker incidence rates. Table 11 from Filkins, Mortimer, Post, and
Chapman (1973) provides an apt illustration of the problem.

These data, based on a sample of 709 DUI offenders from three ASAP sites,
nevertheless provide some indication of the percentage of the convicted DUI population
whose drinking patterns deviate from social use levels.

Vingiles (1983) reached similar conclusions in her extensive review of the literature
on DUI drinking status classification. The DUI offenders classified as problem drinkers
ranged from 2 percent to 89 percent across the various studies. Vingilis estimated that
30-50 percent of DUI offenders would most likely be alcoholic.

Sutker, Brantley, and Allain (1980) evaluated the MMPI profiles of 500 DUI of-
fenders, all of whom were found to share mild antisocial tendencies. Four profile patterns
were identified and were found to differ significantly on levels of self-reported drinking.
Profile groups also differed significantly in race, age, and education. The authors
reported a strong association between elevated levels of self-reported drinking and
_ patterns in which indices of depression and social deviance were also elevated. Compar-
ing profile patterns of DUI offenders with those of alcoholics and psychiatric patients
revealed only a modest overlap among the groups.

Fine, Scoles, and Mulligan (1975) used clinical rationale to develop a three-group
classification typology for 1,500 DUI first offenders. The three groups were differen-
tiated primarily by quantity, frequency, and circumstances of alcohol consumption.

59
BACKGROUND PAPERS

Table 11. Percentages of drivers classified into three drinker categories by .
various classification methods _

Scale - Social drinkers Excessive drinkers Problem drinkers

 

MF questionnaire 62 ' 26 a 12
MF interview 71 06 - 22
Total MF 64 19 17
CRIT’ 17 29 53
Presentence investigation 52 30 18

Psychometrist 46 . 21 33

 

* Combined criterion consisting of driver record and medical and social agency records.

Struckman (1975) grouped DUI offenders into four categories: social drinker, prob-
lem drinker, serious problem drinker, and chronic alcoholic. This drinker-type diagnosis
was based on information from a number of sources, including driving record, criminal
arrest record, Mortimer-Filkins test, and interviews. The reliability of this classification
strategy was quite good, but no significant recidivism differences were found for control
group subjects classified into drinker types.

Home! (1980) developed a typology based on biographic, demographic, driver record,
and criminal record data. Homel hypothesized the existence of six operationally
anchored groups of convicted DUI offenders: never-convicted-again drivers, minor
motoring offenders, serious motoring offenders, dedicated drinking drivers, criminal
offenders, and drive-disqualified offenders. The description of group differences based
on measures such as marital status, age, occupational status, income, BAC, driver record
entries, and response to penalties provided a logical characterization of subtypes within
the DUI offender population. No attempt was made to substantiate the hypothesized
typology statistically, either by comparison of mean differences for cases classified using
this typology, or by cross-validation on an independent sample of DUI offenders.

Formal multivariate taxonomies. Only a small number of DUI-offender studies
employing formal methods of factor and cluster analysis have been reported in the
literature, and they are of relatively recent origin. Among the earliest was a study by
Steer, Fine, and Scoles (1979). These authors applied a hierarchical cluster analysis
technique to three indices of drinking status (BAC and quantity-frequency indices) and
psychoneuroticism scores collected from 1,500 first and repeat DUI offenders. The
analysis produced a number of types, but the resultant hierarchical structure was
complex and difficult to interpret. The authors therefore resorted toa simpler procedure
of forming 16 clusters from a binary mean split of the four measures (2* =16). The seven
most predominant subtypes, containing 87.6 percent of the cases, were found to differ
significantly on several relevant external variables: ethnicity, number of prior DUIs, prior
treatment for alcoholism, prior drug use, and father’s alcohol use.

Scoles, Fine, and Steer (1984) classified 124 non-DUI high-risk drivers using the
Sixteen Personality Factor Questionnaire (16PF). Through Modal Profile Analysis, 91.1
percent of the drivers were assigned to one of seven types, named on the basis of the
largest high score trait: intelligent, shrewd impulsive, shrewd controlled, warmhearted
resourceful, warmhearted adventurous, assertive, and resourceful. The authors con-
cluded that the 124 high-risk drivers were easily able to distort their responses on the
16PF in a socially desirable fashion, and that the utility of the 16PF for assessing
personality disturbance within the high-risk driver population must be questioned.
EPIDEMIOLOGY AND DATA MANAGEMENT

Donovan and Marlatt (1982) identified five subtypes through hierarchical cluster
analysis of 17 driving-attitudinal, personality, and hostility measures from 172 DUI

offenders. These five subtypes were externally validated through comparison on —

demographic, drinking, arid driving-risk variables. Subtypes 2 and 5 were less deviant
than the other three subtypes. Subtype 2, in addition to being the largest group, also
presented the highest overall level of emotional adjustment. Subtype 2 members had the
lowest levels of driving-related aggression, depression, sensation seeking, and overt and
covert hostility. Subtype 5 members had slightly higher scores on these dimensions, but
the remaining three subtypes expressed much more deviant levels of these risk-
enhancing characteristics. Subtypes 1 and 4 were found to have particularly high levels
of risk-enhancing traits (e.g,, high levels of driving-related aggression and hostility, and
low levels of assertiveness, perceived control, and emotional adjustment). The remaining
subtype (subtype 3) was characterized by high levels of depression and resentment and
low levels of assertiveness and emotional adjustment.

Donovan, Queisser, Umlauf, and Salzberg (1984) continued investigating these per-
sonality subtypes through analysis of subsequent 3-year driving records. Subtype mem-
bership was not a significant predictor of DUI recidivism or accidents. However,
significant differences were found for other violation types.

Wells-Parker, Cosby, and Landrum (1985) used an inverse factor-analytic procedure
(Q-mode factor analysis) to develop a typology of 353 DUI offenders who were referred
to a probation and rehabilitation program in Mississippi. The variables used in the
clustering consisted of 45 measures representing different types of traffic- and criminal-
offense information available from driver and criminal record files.

The cluster analysis resulted in five subgroups that the authors characterized as
follows: low (overall) offense group, mixed (offense) group, traffic (moving violations)
group, public drunkenness group, and license offense (equipment and licensing viola-
tions) group. The classification accuracy of the cluster groups was verified through a
multiple-discriminant-function procedure. Incorrectly classified individuals were moved
to the group indicated by the discriminant function. There was an overall agreement of
84.4 percent between the two classification procedures.

The authors cross-tabulated the typology against several external measures and found
statistically significant relationships on the majority of the comparisons. In most
instances, the relationships were intuitively plausible. Of particular interest were
relationships with BAC, Mortimer-Filkins scores, drinking status, and subsequent 24-
month accident and DUI recidivism rates. The public drunkenness and license groups
had the highest percentage of offenders with Mortimer-Filkins scores in the problem
drinker range (24 percent versus 5 percent for all groups combined). The public
drunkenness group also had the highest BAC levels, and by far the highest rate of
subsequent accidents. The license group and public drunkenness groups had the highest
DUI recidivism rates. The low offense, mixed, and traffic groups had comparatively
lower proportions of problem drinkers, and lower rates of recidivism. The low offense
and mixed groups also had the lowest accident rates, in contrast to the traffic group,
which had the second highest accident rate of the five types. These three groups were
substantially younger than the other two, and the traffic group was youngest of the five
(mean = 33.1 versus 42.8 for groups 4 and 5 combined).

The authors compared the characteristics of their typologies with those of Steer, Fine,
and Scoles (1979) and Donovan and Marlatt (1982). Although the resulting typologies
reflect a number of dissimilarities, all three exhibited types that varied in terms of
problem-drinker status (severity), age, and the extent to which the driving record reflects
a general disregard for traffic laws (elevated moving violation and accident rates).

Sacramento DUI Offender Typology Study. Probably the largest multivariate typol-

61
62

_BACKGROUND PAPERS

ogy study of DUI offenders was that of Arstein-Kerslake and Peck (1985). The primary
objectives of this study were to:

1. develop and cross-validate DUI offender typologies based on psychometric and ,
_ nonpsychometric variables and . -

2. assess the extent to which DUI recidivism and DUI treatment-program com-
pliance can be predicted from traffic safety, criminal record, demographic, and
psychometric variables. :

These analyses were performed on data from 7,316 DUI offenders initially collected
during the operation of the California Driving Under the Influence (CDUI) project in

" Sacramento, California, from September 1977 through January 1981 (Reis 19822, b). The

Reis analyses focused on the question of the relative effectiveness of various randomly
assigned countermeasures and did not address the data from classification or prediction
perspectives. ‘

' Two sets of variables were used as scores in constructing first-offender, repeat-

offender, and total sample typologies: the psychometric domain and the descriptive
(nonpsychometric) domain.

The psychometric variables were:
Conforming compliance/acting-out aggressiveness
Extroversion/introversion
Sanguine, self-confident/anxious, depressed
Moralistic, conservative/nontraditional, unconstrained
Paranoid-suspicious/naive trust
Residential stability
Alcohol consumption/quantity-frequency
Alcohol problems -
Physical health problems
Treatment receptiveness
Financial status, employment situation
Familial interaction, living situation
Social interaction and involvement
The descriptive variables were:
Age
Average BAC at arrest
Intake diagnosis —a clinical rating of drinking problem severities
Client attitude at intake
Average monthly income
Educational level
Marital status
Number of marriages
Number of dependants
Occupational socioeconomic status
Intake test score — standardized or modified Mortimer-Filkins test
Traffic conviction record — moving and nonmoving
Traffic accident record
EPIDEMIOLOGY AND DATA MANAGEMENT

Alcohol-related accidents and convictions
Criminal record entries

The psychometric variables are scales on the Life Activity Inventory (LAI) and were
selected because their factorial structure and psychometric properties have been well-
established (Reis 1982a, b). Both the psychometric and descriptive variables were
standardized before performing the cluster analysis.

It should be noted that the descriptive domain variables were available for the entire
sample, whereas the measures in the psychometric LAI domain were only obtained on
the subsample of 2,889 assigned to followup interview conditions in the Reis study.

A K-means cluster program identified nine clusters from analysis of the psychometric
variable domain. These DUI offender types were characterized on the basis of their
average scores on both psychometric and nonpsychometric variables. Descriptions of
the nine clusters follow. ,

e Negligent Operator (cluster 1): Members express behaviors commonly as-
sociated with negligent operators (youthfulness, acting-out aggressiveness,
etc.). Levels of social interaction are high, and levels of residential stability and
financial status are low. This group had the highest rate of alcohol-related and
total accidents and the second highest rate of moving traffic violations. They
also had the highest rate of drug arrests and the lowest rate of program
compliance.

e Pre-DUI Alcoholic I (cluster 2): This has a strong, alcohol-related component,
but not as extreme as cluster 3. This group had high level of mistrust, higher
than average conservatism and depression, and high levels of aggressiveness. It
also had a very high percentage of minorities (55 percent) and the highest rate
of moving, nonmoving, and reckless driving convictions. The group was
predominantly composed of first offenders (63 percent).

e DUI-Alcoholic (cluster 3): A number of measures indicate an extreme alcohol
problem and a high traffic-safety risk. This group had the highest levels of
anxiety-depression, and aggressiveness, and the highest quantity frequency
index. It also had high levels of health problems.

e Pre-DUI Alcoholic II (cluster 4): This cluster is descriptively similar to cluster
2, although these two cluster groups do differ with respect to psychometric
measures. Cluster 4 expressed high levels of introversion and isolation and high
levels of lack of constraint.

e “Mid-Life Crisis” Problem Drinker (cluster 5): Members report high levels of
stress in interpersonal relationships, high levels of physical health problems,
high unemployment, and high average age. A relatively large number perceived
alcohol as a problem in their lives despite having the lowest Q-F index.

e Deceptive Problem Drinker (cluster 6): Levels of socially desirable attributes

were surprisingly high, and levels of improbable response were very high as
measured by the “lie” scale.

e White-Collar Controlled Problem Drinker (cluster 7): A relatively high per-
centage of persons employed in white collar occupations appear to be control-
ling their drinking and pose a relatively low traffic safety risk. This cluster

exhibited high levels of conforming-compliance, self-confidence, and trust, and
a comparatively low Q-F index.

e Blue Collar Controlled Problem Drinker (cluster 8): Members possess at-
tributes similar to cluster 7, except for their generally lower level of
socioeconomic status.

e Social-Normative Problem Drinker (cluster 9): Age distribution is similar to
clusters 2 and 4, but cluster 9 members express much higher levels of

63
64

BACKGROUND PAPERS

socioeconomic status and socially desirable attributes. It may well be that these
persons tend to consume excessive amounts of alcohol in settings in which such ©
behavior may not be considered deviant (i.e., socially normative).

Statistically significant differences in subsequent 4-year accident rate, traffic convic-
tion rate, and DUI recidivism were found among the nine psychometric clusters.

A separate K-means cluster analysis was performed on nonpsychometric variables
(driver record, criminal record, and intake interview). Descriptions of the 10 clusters

follow:

Cluster 1: Lowest average age (22.0 years); lowest average BAC at arrest; .
one-fourth without high school diploma; high percentage of males (92 percent);
relatively low percentage of persons married or cohabitating; highest number
of moving and nonmoving violations; 67 percent first offenders.

Cluster 2: Average age 24.0 years; relatively high levels of educational attain-
ment and white collar employment; lowest average income; average levels of
acting-out aggressiveness; exclusively unmarried; 82 percent male; 74 percent
first offenders. ,

Cluster 3: Average age 24.2 years; one-fourth without high school diploma; 91
percent male; highest average score on accident composite measure and drug-
alcohol composite measure; high levels of alcohol problems and treatment
receptivity; 58 percent multiple offenders.

Cluster 4: Average age 30.4 years; highest percentage of females (32 percent);
no never-marrieds; 78 percent either separated, divorced, or widowed; highest
percentage of first offenders (78 percent).

Cluster 5: Average age 33.2 years; smallest descriptive cluster; above-average
incidence of accident and convictions; high percentage of minorities (36 per-
cent); very high levels of hostility and suspicion as judged by diagnostic coun-
selor; 56 percent first offenders.

Cluster 6: Average age 34.6 years; above average income; high percentage of
blue-collar-workers; highest number of dependents; highest percentage cur-
rently married (84 percent); highest percentage of minority members (37
percent); 61 percent first offenders.

Cluster 7: Average age 35.1 years; above-average incidence of accident and
convictions; very high levels of criminal record entries; contains all persons who
refused the BAC test associated with entry arrest (approximately 50 percent of
cluster 7 members); highest proportion unemployed (42 percent); largest
descriptive cluster; 59 percent multiple offenders.

Cluster 8: Average age 35.7 years; very high score on alcohol problem severity
on psychometric inventory administered at intake (refers to standardized score
on Mortimer-Filkins or CDUI scale —both tests were used at different times
during the operation of the CDUI project); high levels of acting-out aggressive-
ness and depression; high levels of perceived alcohol problems and physical
health problems; high levels of treatment receptivity, lowest level of satisfaction

in marriage or marriagelike relationship; highest percentage of multiple of-
fenders (71 percent).

Cluster 9: Average age 38.6 years; highest occupational status (83 percent
white-collar — professional/technical, management/administration, sales); high
levels of educational attainment (41 percent 4 or more years of college); high
levels of marital stability and familial and social interaction; low levels of
acting-out aggressiveness; 62 percent first offenders.

Cluster 10: Average age 53.0 years; no members under 30 years old; 95 percent
of members over 40 years old; relatively high unemployment (39 percent);
EPIDEMIOLOGY AND DATA MANAGEMENT

relatively low average income; lowest levels of educational attainment (49 per-
cent with fewer than 12 years education); high percentage of females (24 per-
cent) and minorities (30 percent); 50 percent married more than once;
46 percent currently divorced or widowed; lowest levels of acting-out aggres-
siveness; high levels of introversion and physical health problems; low levels of
familial and social interaction; 64 percent multiple offenders.

Comparisons of the descriptive clusters on subsequent 4-year driver record indicated
statistically significant differences in fatal accident rates, accidents involving alcohol,
moving traffic convictions, and DUI recidivism. Clusters 1, 3, 5, and 7 tended to have the
poorest records, whereas cluster 9 consistently had the lowest number of driver record
entries.

Both the psychometric and descriptive (nonpsychometric) solutions were cross-
yalidated using a 25-percent subsample not used during the information of the cluster
solution. In addition, the cluster solutions were submitted to discriminant function
analysis. In each case, differences among clusters based on relevant clustering variables
were substantial, resulting in 89.9 percent correct classification for members of the
psychometric cross-validation subsample and 67.6 percent correct classification for
members of the descriptive cross-validation subsample. For the psychometric cluster
solution, 86.8 percent of the variance in cluster membership was explained by five
discriminant functions. For the descriptive cluster solution, five functions accounted for
80.2 percent of the variance in cluster membership.

Since Arstein-Kerslake and Peck performed separate cluster analyses within each
domain, the resultant typologies did not constitute a single integrated system. They were
able, however, to investigate the. structural relationship between the two by cross
tabulating the two systems. The end product of this analysis was a contingency table
showing how membership in the psychometric domain clusters were distributed across
cach category of the descriptive variable typology. The results are summarized in table 12.

Although the relationship between the two typologies was significant, the association
was quite low (Cramer’s v=0.22, p<0.001). Thus, the two systems are much more
independent than they are overlapping. :

It would be possible to view table 12 as a two-dimensional taxonomic system if one of
the dimensions can be viewed as subordinate to the other. For example, if one considers
the psychometric typology to have logical precedence over the descriptive taxonomy,
then one could view the groups of the former as types and those of the latter as subtypes.
To illustrate from table 12, consider the DUI alcoholic type (type 3). We find that 64
percent of this group “maps” into descriptive subtypes 7 and 8. Hence, these combina-
tions could be numerically coded as 3.7 and 3.8. Both subtypes 7 and 8 clearly reflect
serious alcohol problems and the psychosocial manifestation of alcoholism. Therefore,
they corroborate the psychometric taxonomy (DUI alcoholic) but provide some addi-
tional differentiation. Subtype 7 manifests more antisocial traits (criminal record),
resistance to authority (implied consent chemical test refusal), bad driving record
(accidents, moving violations, prior DUIs), and high unemployment. Subtype 8 appears
to have the most acute alcohol problem but the members appear more receptive to
treatment and to be aware of their problems.

Arstein-Kerslake and Peck point out that the clusters vary greatly in terms of their
similarity/dissimilarity and that many of the clusters would not be fixed in time. Intuitive-
ly, one would expect flux with persons changing clusters as they age and experience
deterioration or improvement in their drinking status and its relation to driving. For
example, it is known that some young people pass through a period of excessive drinking
and use of alcohol in conjunction with driving. Such people might not be labeled social

drinkers, even though many will eventually leave the drinking-driver population and
never progress to alcoholism. A longitudinal repeated-measures factorial study would
be required to validate the hypothesized transitions empirically .
BACKGROUND PAPERS

Table 12. Cross tabulations of correspondence between psychometric cluster membership
and descriptive cluster membership ,

 

 

Count Psychometric clusters

Row percent ; Row
Column percent 1 2 3 4 5 6 7 8 9 total
1 .37 25 10° #427 3 21 9 6 39 177

20.9 14.1 5.6 15.3 17 8611.9 51. 34 220 11
106 13.7 3.2 72 14 5.4 2.6 2.4 85

2 86 29 21 74 16 =. 54—s8B 3 115 406
21.2 71 52 182 39 133 2.0 0.7 283 141
247 158 6.7 19.7 73 138 2.3 12 25.2

3 48 9 26 47 14 B 14 8 39° = 218
22.0 4.1 119 216 6.4 6.0 6.4 3.7 17.9 75
13.8 49 8.3 12.5 6.4 3.3 4.0 3.2 8.5

4 39 25 16 41 19 37 20 11 16 304
12.8 8.2 5.3 B5 63 188 6.6 36 25.0 10.5
11.2 8 13.7 5.1 10.9 86 145 5.7 4.4 16.6 |

5 9 7 5 13 8 9 13 7 13 84
10.7 8.3 6.0 15.5 9.5 10.7 15.5 8.3 15.5 2.9
2.6 3.8 1.6 3.5 3.6 23 3.7 2.8 2.8

6 9 9 15 12 8 39 85 63 9 249
3.6 3.6 6.0 48 3.2 15.7 341 253 3.6 8.6
2.6 49 48 3.2 3.6 99 244 25.3 2.0

7 96 4 11 94 71 95 nN 65 81-735
Bi 61 61 128 05 29 97 88 NO 25
276 4246 «3520281350 24.2 203 261177

8 16 18 89 19 28 5 20 36 23 254
63 71 35.0 75 11.0 2.0 19 14.2 9.1 8.8
46 98 283 5.1 12.7 18 5.7 14.5 5.0

9 2 3 15 14 4 42 100 29 49 259
12 1.2 58 5.4 15 16.2 38.6 112 189 9.0
0.9 16 48 3.7 18 10.7 28.7 11.6 10.7

10 5 B 7 34 43 57 9 21 13 202
2.5 6.4 3.5 168 213 28.2 45 10.4 6.4 7.0
1.4 7A 2.2 9.1 19.5 145 2.6 8.4 28

Column total 348 183 315 375 220 392 349 249 457 2,888
12.0 63 10.9 13.0 7.6 13.6 12.1 8.6 15.8 100.0

 

Chi-square = 1162.3 with 72 df. significance = 0.0000
Eta (descriptive cluster dependent) = 0.35
Cramer's lambda = 0.22
EPIDEMIOLOGY AND DATA MANAGEMENT

It is instructive to note that Arstein-Kerslake and Peck failed to uncover any group
labeled social drinkers. The following quotation illustrates their reasoning.

“Hapless social drinker” does not constitute a sizeable enough portion of the
DUI offender population to be identified as a separate subtype. In an
intuitive sense, there are three dimensions which seem to characterize the
differences between the nine psychometric clusters: (1) consumption of
alcohol (moderate to excessive), (2) problem drinker predisposition (tran-
sient to chronic), and (3) negligent operator characteristics (none too many).
Different weightings on these dimensions for each cluster contribute to the
differential accident/conviction levels among clusters. Even those clusters
with low accident/conviction levels (e.g., very low negligent operator charac-
teristics) have high enough levels on other dimensions (e.g., problem drinker
predisposition) to preclude their being classified as “social drinkers.”

The above conclusion requires clarification and tempering. Each cluster is really an
aggregation of people into averages, and some of the clusters imply relatively moderate
levels of alcohol consumption. Individual variation may be substantial within groups and
latent subtypes that were too small in number to emerge as a distinct type. Obviously
some of those among convicted DUI offenders could be characterized as social drinkers.
However, as Stewart., Epstein, Gruenewald, Laurence, and Roth (1987) point out, the
very term “social drinker” is imprecise and of dubious scientific value. It can be more
meaningful to talk about the amount, frequency, and pattern of alcohol consumption,
and about points on this continuum where problems are likely to occur.

Discussions and Conclusions

The preceding review of the literature on DUI offender characteristics indicates that
convicted offenders differ from the general driving population on a wide range of
variables. Although they share some of the same characteristics as problem (negligent)
drivers, alcohol-involved accident drivers, and alcoholics, the overlaps are not large with
any of the three in an absolute statistical sense. The convicted DUI offender represents

acombination of the traits of all three, plus a substantial amount of unique DUI-offender
characteristics.

The research on multivariate typologies and other taxonomic systems indicates that
the DUI-offender population contains some distinct subgroups, or types, and there is
some consistency across studies in the structure of the taxonomies and the subgroups
resulting from them. These classification systems.may be useful in providing insight

concerning etiology of problem drinking (driving) and in suggesting potentially effective
modes of treatment. .

The concept of the impaired problem driver has merit and is consistent with some of
the typologies described above. To a limited extent, DUI recidivism can be predicted,
and the characteristics of recidivists clearly indicate a profile of persons with progres-
sively increasing drinking problems and negligent-driving problems. Nevertheless, the
results of Arstein-Kerslake and Peck (1985) suggest that the majority of first offenders
are statistically indistinguishable from repeat offenders and represent equally high
accident risks. This finding suggests that most first offenders are problem-drinker drivers
who have simply not yet had their second DUI offense.

Current Issues and Problems

The major problems and questions in this area are identified and briefly discussed in
this section. Extensive discussion of these topics is unnecessary since they follow imme-
diately from the literature review presented above. In addition to the major headings

67
68

BACKGROUND PAPERS

used in the literature review, a subsection focused on the data management aspects of

the problem is included below.

Alcohol Involvement in Crashes

The most fundamental problem in establishing alcohol involvement in highway
crashes stems from the incomplete testing and reporting of BAC. data. Enormous
progress has been made during the past decade in obtaining BAC data from fatal crashes
through FARS, by means of which the States are expected to report all relevant

information to NHTSA. Although a few States report BAC data for all fatal crashes

within their borders and approximately one-third of the States report the BAC in at least
85 percent of their fatal crashes, most States fall far short of acceptable reporting.
standards for this crucial bit of information from each fatal crash. Overall, BAC data are
now reported to FARS for 74 percent of fatally injured drivers, but for only 45 percent
of the surviving drivers involved in fatal crashes. Thus, the shortfall of data for this latter
category of driver represents another major problem. .

Far fewer BAC data are available for surviving drivers injured in highway crashes,
primarily because American hospitals generally will not release such information.
Consequently, estimates of the legally impaired drivers injured in highway crashes range
from 20 percent to 40 percent, with the best estimate being around 25 percent. This
question is currently being addressed in Ontario, Canada, in a research project funded
by NIAAA (principal investigator: Evelyn Vingilis).

Data on BAC are especially scarce for crashes with property damage only, i-e., those
collisions involving reportable damage to property but none to humans. Although these
crashes are of less importance for epidemiologic research than the two previous
categories involving human injury, it would nevertheless be useful to obtain BAC data
across the full continuum of reportable crashes for the sake of fully explicating the
relation of alcohol amount to the relative seriousness of the crash.

Alcohol and Noncrash Drivers

The traffic safety community is experiencing a shortfall of up-to-date information
concerning the population-at-risk, i.e., drivers who are not involved in crashes. Although
BAC data are the most important, it is also necessary to obtain additional information
concerning age, gender, purpose of trip, perceived risk of being stopped at an enforce-
ment checkpoint, perceived risk of being arrested for drunk driving, and so forth. Such
data are crucial for evaluating the progress and current status of both public health and
public safety programs. As noted above, NIAAA is currently supporting a research
project focused on alcohol tolerance among drinking drivers that is obtaining substantial
information about the nocturnal population at risk (principal investigator: M.W. Per-
rine). Data from this 5-year roadside survey of 42,000 drivers will be reported as obtained
over the course of the next 4 years of field activity.

Characteristics of Drunk Drivers

The present state of research knowledge already permits differentiation within the
category of drunk driver. Indeed, recent research provides for an increasing number of
differentiations not only among drunk drivers, but also among drinking drivers, as
described by Perrine (1987). However, further increases, refinements, and validations
of such differentiations among drinking drivers are necessary in order to address the
major problems in this area more effectively. The specific shorter term goals for such
efforts should be:

e early identification of potential DUI offenders through increased
differentiation among drinking drivers,
EPIDEMIOLOGY AND DATA MANAGEMENT

e more accurately targeted sanctions for convicted DUI offenders,

‘e more accurately individualized referrals of DUI offenders from the
courts to selected treatment programs available in the locality,

e more individually customized counseling and treatment programs for
DUI offenders,

e increased success rates in terms of alcohol treatment program
compliance and completion, and

e lower recidivism rates.

Although these goals are focused initially on specific deterrence and the convicted DUI
offender, the broad longer term goals involve aspects of both general deterrence and
public education (Perrine 1987).

The NIAAA is currently supporting a research program designed to address the
above goals through grants concerned with the probabilities of drunk driving among the
USS. public and among convicted DUIs (AA06774 and AA06926, principal investigator:
M.W. Perrine). This research program is based upon 15 concurrent, interrelated projects
that focus on five different but interdependent segments of the American drinking and
driving public: the general driving population, the nocturnal driving population, the
convicted DUI first-offender population, the convicted DUI multiple-offender popula-
tion, and those arrested for DUI, but not convicted. The basic rationale for this approach
derives from analyzing the known characteristics of those in these interrelated popula-
tions to determine the similarities among those who “get into trouble with alcohol,” as
well as the differences between those who do and those who do not (Perrine 1987). The
results of this ongoing research program should enable developing much more specific —
and thus much more effective—means for prevention and intervention in this major
public health problem area.

Data Sources and Management

Much epidemiologic research on drunk driving (as in many other areas) depends on
data collected by third parties and available in the form of official records. Thus, the
adequacy, utility, and validity of such research are seriously constrained by the accuracy
and completeness of the data themselves, as well as by the data collecting and data
reporting. The systems for DUI processing (from the point of arrest through the
monitoring of adherence to the imposed sanctions) vary greatly from State to State, but
apparently all such systems are prone to loopholes and failures. Indeed, it is necessary
to analyze the individual State DUI processing systems carefully, not only to understand
their functioning, but especially to identify points for dropouts and other failures. Only
by so doing can the accuracy and completeness of the resulting DUI data be assessed.

An extensive evaluation of the drunk-driving countermeasure system in the State of
California has recently been completed, funded by the NHTSA. The first of the eight-
volume series is concerned with an analysis of DUI processing from arrest through
postconviction countermeasures (Perrine 1984). The objectives of the project were to
develop process flow charts and a description for the whole DUI system, as well as to
identify sources of system inefficiency or modes of circumvention of specific provisions
of the laws and the system. The interorganizational task force formed to accomplish these
objectives represented all major constituencies in the DUI countermeasure system: law
enforcement agencies; prosecutors; municipal, superior, and juvenile courts; pro-
gram/service providers; State and county alcohol-program administrators; probation
officers; and the Department of Motor Vehicles (DMV).

A subsequent study was conducted to identify deficiencies in the California DUI
countermeasure system and to evaluate empirically the frequency with which DUI
offenders avoid timely processing or circumvent system countermeasures owing to these

69
70

BACKGROUND PAPERS

deficiencies (Helander 1986). To accomplish these objectives, a sample of DUI of-
fenders was tracked all the way through the system in order to describe and analyze the

flow of the system. \

A total of 3,959 DUI offenders arrested by 44 law enforcement agencies in seven
sample counties was tracked through the DUI system from the point of arrest through
postconviction countermeasures. A separate sample of 701 convicted DUI offenders
referred to alcohol education/treatment programs in the seven sample counties was
identified from program provider records and tracked through DMV, court, and pro-
gram records. The principal results were:

_e@ Probability of conviction for a DUI offender varied widely, depending on the
county and court in which the offense was adjudicated. The use of sanctions
also varied widely by county and court.

e Most alcohol education/treatment program dropouts were not reported to the
DMV by the courts, and a substantial percentage of DUI offenders avoided
license suspension as a result.

e Nine percent of drivers arrested for DUI were under license suspension or
revocation at the time of arrest. Only 20 percent of these drivers were convicted
for the offense of driving while license was suspended or revoked.

e Aszurprisingly large percentage of DUI offenders was unlicensed or had more

_ than one driver record, that is, they had multiple licenses under different names
or errors in the files (e.g., wrong name or date of birth) created additional
records that were not charged to the driver.!

Based on the study findings, Helander (1986) concluded that:

e The probability of punishment for DUI offenses must be increased in order to
produce any large-scale impact on the problem of drinking and driving.

e Thecitation and conviction rates of those who drive while suspended or revoked
must be improved ‘if license suspension is to remain an effective and credible
traffic safety countermeasure.

e Ifthe DUI countermeasure system is to function as a true system, goals and
objectives must be developed along with a management information system to
assess the achievement of those goals and objectives.

e Improvement is needed in the accuracy of records in the DUI countermeasure
system.

The final volume in this series evaluating the California drunk driving countermeasure
system consists of an overview of study findings and policy recommendations prepared
by Peck (1987). Among many other important findings, Peck emphasized the need to
improve the management information system and monitor its quality control through
periodic process evaluations. For example, it is necessary to monitor:

e time lags in the processing system;
e characteristics of the plea bargaining process, as well the plea
bargaining rate;

e the rate of dismissal of prior DUI convictions;

1 In California, all traffic convictions for moving traffic violations and for major violations, such as drunk
driving, are reported to the DVM and placed on the driver's driving record file. The driving record also
includes fatal and injury accidents and property damage accidents involving more than $50 to a vehicle. The
driver record entries are retained from 3 to 10 years, depending on the nature of the offense . Other States
have similar file systems, although the specific provisions and retention periods will vary from State to State.
EPIDEMIOLOGY AND DATA MANAGEMENT

e the rate of compliance with imposed sanctions;

«the rate of completing the sanctions (especially the
treatment/education programs);

«the incidence of jury trials; and
+ the incidence of implied consent refusals to submit to a breathtest.

Some of the more serious shortfalls identified in the data system are (Peck 1987):
e accurate reporting of the BAC at the time of arrest;

e accurate determination of the offense status (whether a first, second,
third, or more DUI offense); --

e determining whether a jail sentence was imposed, but more important,
determining whether it was actually served to completion;

e determining whether community services were actually performed

e determining positively through affirmative evidence that the alcohol
treatment program had actually been completed by the DUI offender.

Research Questions

In 1987, as part of a continuing effort by NIAAA to assess research opportunities and
needs in the field, a series of meetings focused on issues in alcohol research on safety
and trauma. A series of papers provided solicited advice on extramural research
priorities (these papers are published in a special issue of Contemporary Drug Problems,
Spring 1988). In addressing the research issues, needs, and opportunities in the area of
alcohol, trauma, and traffic safety, Perrine (1988) formulated a number of specific
questions that should be addressed through epidemiologic and field studies. These
questions are also appropriate to consider in the present context.

e How frequently does driving actually occur after drinking among the US.
motoring public?

e How firm is the linkage between social drinking activities and subsequent
driving activities? ‘

e Why are some drivers involved in crashes after drinking, whereas others are
not —even at the same BACs?

e Towhat extent do fatally injured drivers with high BACs differ from other high
BAC drivers who crash but are not fatally injured, or are not injured at all, or
who are not even involved in a crash?

e To what extent is alcohol involved in crashes, and to what extent is alcohol
responsible for crashes?

A number of research questions and issues were also formulated concerning
idiosyncratic characteristics of drinking drivers (Perrine 1988), namely;

- ¢ What characteristics can be identified to distinguish among the various
groups/types of individuals across the spectrum of drinking drivers?

e Towhat extent is it possible to differentiate drinking drivers who avoid detec-
tion, accidents, and conviction of DUI from drinking drivers who are arrested
and convicted of DUI?

e To what extent is it possible to identify future DUI offenders in advance, that
is, before the fact? At what point or stage of development is such identification
possible?

© Towhat extent is it possible to identify potential DUI reoffenders or recidivists
in advance, for example, after the first DUI offense but before the second? At
what point or stage of development is such identification possible?

71
72

BACKGROUND PAPERS

e Towhat extent do such advance indicators of DUI consist of stable, persistent
characteristics, as opposed to more temporary and transitional aspects? In —
other words, to what extent does a person have a “predisposition”-to become a
DUI offender?

e To what extent do factors other than alcohol contribute to high-risk drinking
and driving and ultimately to alcohol-involyed crashes?

e Isadrinking driver a potential DUI offender when actually alcohol-impaired,
or simply when the BAC is 0.10 or higher?

These questions may prove valuable in designing new research projects to investigate
further the role of alcohol and traffic safety.

Recommendations

A number of recommendations emerge clearly from the foregoing literature review
and examination of current issues and problems. The most important of these recom-
mendations are listed below without further comment.

e Develop policies and procedures to ensure that uniform and consistent alcohol
data are obtained for all highway crashes.

e Develop policies and procedures to ensure that accurate alcohol data are
obtained for commercial motor-vehicle operators using the highways.

e Determine feasibility of gathering accurate data on drivers under 0.10 BAC at
enforcement checkpoints.

e Develop more effective roadside survey policies and techniques to collect
increasingly valid data on drunk driving (e.g., incidence and prevalence,
changes in distribution of alcohol concentration).

e Develop a, central monitoring, record keeping, and reporting capability for
drunk driving data.

e Develop and test a valid, cost-effective surrogate for roadside surveys in order

to evaluate countermeasure programs and to monitor public awareness and
perception of risk. .

e Determine more accurately the characteristics of drunk drivers to facilitate
early identification and counseling, to encourage more accurately targeted
sanctions for convicted DUI offenders, to encourage more customized coun-
seling and treatment programs for DUI offenders, and thereby, it is hoped, to
obtain increased success rates in terms of alcohol treatment compliance and
completion, as well as lower recidivism rates.

REFERENCES

Arstein-Kerslake, G.W., and Peck, R.C.A Typological Analysis of California DUI Offenders and
DUI Recidivism Correlates. Technical Report. Washington, DC: National Highway Traffic
Safety Administration, 1985.

Beitel, G.; Sharp, M.; and Giauz, W. Probability of arrest while driving under the influence of
alcohol. Journal of Studies on Alcohol 36:237-256, 1975.

Borkenstein, R.F.; Crowther, R.F.; Shumate, RP.; Ziel, W-B.; and Zyiman, R. The Role of the
Drinking Driver in Traffic Accidents. Bloomington, IN: Department of Police Administra-
tion, Indiana University, 1964. ;

Borkenstein, R.F.; Crowther, R.F.; Shumate, R.P.; Ziel, W.B.; and Zylman, R. The role of the
drinking driver in traffic accidents. (The Grand Rapids Study). Blutalkohol, 11(Supple-
ment 1):1-132, 1974.
EPIDEMIOLOGY AND DATA MANAGEMENT

Burch, H.E. Preliminary Analysis of the Rehabilitation Countermeasure: Analytic Study for 1973.
Contract No. DOT-HS-161-2-252. Washington, DC: National Highway Traffic Safety
Administration, 1974.

Cahalan, D.; Cisin, I.H.; and Crossley, H.M. American Drinking Practices: A National Study of
Drinking Behavior and Attitudes. New Haven, CT: College and University Press, 1969.

Carr, B.; Borkenstein, R.F.; Perrine, M.W.; Van Berkom, L.C.; and Voas, R.B. International
Conference on Research Methodology for Roadside Surveys of Drinking- Driving. Technical
Report No. DOT HS-801-220. Washington, DC: National Highway Traffic Safety Ad-
ministration, 1974. pp. 1-75.

Denberg, D. Recidivism and Collision Involvement of Impaired Drivers in Ontario. (Substudy No.
587). Toronto: Addiction Research Foundation, 1974.

Donelson, A.C.; Beirness, D.J.; and Mayhew, D.R. Characteristics of drinking drivers. Impaired
Driving. Report No. 1. Ottawa: Department of Justice, Canada, 1985.

Donovan, D.M., and Marlatt, G.A. Personality subtypes among driving while-intoxicated of
fenders: Relationship to drinking behavior and driving risk. Journal of Consulting and
Clinical Psychology 50(2):241-249, 1982.

Donovan, D.M.; Queisser, H.R.; Salzberg, P.M.; and Umlauf, R.L. Intoxicated and bad drivers:
Subgroups within the same population of high-risk drivers. Journal of Studies on Alcohol
46(5):375-382, 1985.

Donovan, D.M.; Umlauf, R.L.; and Salzberg, P.M. Bad drivers: Identification of a target group
for alcohol-related prevention and early intervention. Journal of Studies on Alcohol in
press.

Ellingstad, V.S. An Analysis of Alcohol Rehabilitation Efforts. Analytic Study 6. Contract No.
DOT-HS-045-1-061. Washington, Dc: National Highway Traffic Safety Administration,
1974.

Epperson, W.V.; Harano, R.M.; and Peck, R.C. Final Report to the Legislature of the State of
California in Accord With Resolution Chapter 152, 1972 Legislative Session. Senate
Concurrent Resolution 44-Harmer. Sacramento, CA: Department of Motor Vehicles,
1975.

Evans, L. Fatal and severe crash involvement vs. driver age and sex. 31st Annual Proceedings,

American Association for Automotive Medicine. New Orleans, LA: General Motors Re-
search Laboratories, September 1987. pp. 59-76 -

Fell, J. Alcohol Involvement Rates in Fatal Crashes: A Focus on Young Drivers and Female Drivers.
Technical Report No. DOT-HS-807-184. Washington, Dc: National Highway Traffic
Safety Administration, 1987.

Filkins, L.D.;:Mortimer, R.G.; Post, D.V.; and Chapman, M.M. Field Evaluation of Court
Procedures for Identifying Problem Drinkers. Technical Report No. DOT-HS-031-2-303.
Washington, DC: National Highway Traffic Safety Administration, 1973.

Fine, E.W.; Scoles, P.; and Mulligan, M. Under the influence: Characteristics and drinking practices
of persons arrested the first time for drunk driving, with treatment implications. Public
Health Reports 90(5):424-429, 1975.

Fridiund, G., and Hagen, R. Characteristics of DWI Drivers. Menlo Park, CA: Stanford Research
Institute, 1977.

Greenfield, L.A. Drunk Driving. Special Report, NCJ-109945. Washington, DC: U.S. Department
of Justice, Bureau of Justice Statistics, February 1988.

Hagen, R.E. Effectiveness of License Suspension or Revocation for Drivers Convicted of Multiple
Driving-Under-the-Influence Offenses. Technical Report No. NTIS CAL-DMV-RSS-77-
59. Sacramento, CA: Department of Motor Vehicles, 1977.

Hagen, R.E,; Williams, R.L.; McConnell, E.J.; and Fleming, C.W. An Evaluation of Alcohol Abuse
Treatment as an Alternative to Drivers License Suspension or Revocation. Final Report to
the Legislature in Accord with SB 38 - Gregorio. NTIS No. PB291534. Sacramento, CA:
Department of Motor Vehicles, December 1978.

- Hause, J.; Voas, R.; and Chavez, E. Conducting voluntary roadside surveys: The Stockton
experience. In: Valverius, M.R., ed. Proceedings of the Satellite Conference to the Eighth
International Conference on Alcohol, Drugs and Traffic Safety. Stockholm: The Swedish
Council for Information on Alcohol and Other Drugs (CAN), 1982. pp. 104-113

Helander, C.J. The California DUI countermeasure system: an evaluation of system processing
and deficiencies. Volume 5 of An Evaluation of the California Drunk Driving Counter-

73
74

" BACKGROUND PAPERS

measure System. Technical Report CAL-DMV-RSS-86-97. Sacramento, CA: California
Department of Motor Vehicles, 1986.

Holcomb, R.L. Alcohol in traffic accidents. Journal of the American Medical Association 11 1:1076-
1085, 1938. , oS -

Homel, R. Penalties and the Drink Driver: A Study of One-Thousand Offenders. Volume 1.
Research Report No. 7. New South Wales, Australia: Department of the Attorney —
General, 1980. cot ,

Hurst, P.M. Epidemiological aspects of alcohol in driver crashes and citations. Journal of Safety
Research 5:130-148, 1973. ;

Hurst, P.M. Epidemiological aspects of alcohol in driver crashes and citations. In: Perrine, M.W.,
ed. Alcohol, Drugs, and Driving. Technical Report DOT HS-801-096. Washington, DC:
National Highway Traffic Safety Administration, 1974. .

Hurst, P.M. Blood alcohol limits and deterrence: Is there a rational basis for choice? In: Moskowitz,
H., ed. Alcohol, Drugs, and Driving: Abstracts and Reviews 1(1-2):121-130, 1985.

‘Jelinek, E.M. The Disease Concept of Alcoholism. Highland Park, NJ: Hillhouse Press, 1960.

Lehman, R.J.; Wolfe, A.C.; and Kay, R.D. A Computer Archive of ASAP Roadside Breathtesting

Surveys, 1970-1974. Technical Report DOT-HS-801-502. Washington, DC: National
Highway Traffic Safety Administration, 1975. ,

Levy, P.; Voas, R.; Johnson, P.; and Klein, T.M. An evaluation of the Department of
Transportation’s alcohol safety action projects. Journal of Safety Research 10(4):162-176,
1978.

Lucas, G.W.H.; Kalow, W.; McColl, J.D.; Griffith, B.A.; and Smith, H.W. Quantitative studies of

_ the relationship between alcohol levels and motor vehicle accidents. In: Proceedings of the
Second International Conference on Alcohol and Road Traffic. Toronto: Garden City
~ Press Cooperative, 1955.

Maisto, S.A.; Sobell, L.C.; Zelhart, P.F.; Conners, G.J.; and Cooper, T. Driving records of persons
convicted of driving under the influence of alcohol. Journal of Studies on Alcohol 40(1):70-
77, 1979.

McCarroll, J.R., and Haddon, WJ.A controlled study of fatal automobile accidents in New York
City. Journal of Chronic Diseases 15:811-826, 1962. ,

McConnell, E.J., and Hagen, R.E. Design and Evaluation of a Crash Prediction Strategy. Report
No. 76. NTIS No. PB-81-142671. Sacramento, CA: Department of Motor Vehicles, 1980.

McGuire, F.L. An Evaluation of the Orange County Alcohol Traffic Safety Project. Irvine, CA:
University of California, 1975. .

Miles, W.R. Alcohol and motor vehicle drivers. In: Proceedings of the Thirteenth Annual Meeting
of Highway Research Board. Washington, DC: Division of Engineering and Industrial
Research, National Research Council, 1934.

Moskowitz; H.; Walker, J.; and Gomberg, C. Characteristics of DWIs, Alcoholics and Controls.
Prepared for California Department of Alcohol and Drug Abuse. Los Angeles, CA:
University of California, 1979.

National Highway Traffic Safety Administration. Alcohol and Highway Safety 1984: A Review of
the State of the Knowledge. Technical Report No. DOT-HS-806-569. Washington, pc:
Supt. of Docs., USS. Govt. Print. Off, 1985.

National Highway Traffic Safety Administration. Drunk Driving Facts. Washington, DC: National
Center for Statistics and Analysis, US Department of Transportation, 1987a.

National Highway Traffic Safety Administration. National Accident Sampling System 1985: A
Report on Traffic Accidents and Injuries in the United States. Publication No. DOT-HS-
807-074. Washington, DC: US Department of Transportation, 1987b.

National Highway Traffic Safety Administration. The Economic Cost to Society of Motor Vehicle
Accidents. 1986 Addendum. Washington, DC: US Department of Transportation, Office
of Plans and Policy, 1987.

National Highway Traffic Safety Administration. Fatal Accident Reporting System 1986: A Review
of Information of Fatal Traffic Accidents in the United States in 1986. Publication No.
DOT-HS-807-245. Washington, DC: US Department of Transportation, 1988.

National Institute on Alcohol Abuse and Alcoholism. Drinking and driving. Alcohol Health and
Research World 7(1), 1982.

Nationat Institute on Alcohol Abuse and Alcoholism. Alcohol, other drugs, and safety. Alcohol
Health and Research World 9(4), 1985.

National Institute on Alcohol Abuse and Alcoholism. Siah Special Report to the U.S. Congress on
EPIDEMIOLOGY AND DATA MANAGEMENT

Alcohol and Health. DHHS Publication No. (ADM) 87-1519. Washington, DC: USS.
Department of Health and Human Services, 1987.

National Research Council. Zero Alcohol and Other Options. Limits for Truck and Bus Drivers. _

Special Report 216. Washington, DC: Transportation Research Board, 1987. .

Nichols, J.L., and Reis, RE. One Model for the Evaluation of ASAP Rehabilitation Effort. Technical
Report No. DOT-HS-801-244. Washington, DC: US Dept of Transportation, NHTSA,
1974.

Peck, R.C. An overview of study findings and policy implications. An Evaluation of the California
Drunk Driving Countermeasures System, Final Vol. Technical Report CAL-DMV-RSS-
87-112. Sacramento, CA: California Department of Motor Vehicles, 1987.

Peck, R.C., and Kuan, J. A statistical model of individual accident risk prediction using driver
record, territory and other biographical factors. Accident Analysis and. Prevention
15(5):371-393, 1983. -

Perrine, M.W. Methodological considerations in conducting and evaluating roadside research
surveys. Technical Report, DOT HS-800-471. Washington, DC: National Highway Traffic
Safety Administration, 1971.

Perrine, M.W. Alcohol and highway safety (Chapter 6). In: Second Special Report to the U.S.
Congress on Alcohol and Health. DHEW Pub. No. (ADM) 75-212. Washington, DC:
Supt. of Docs., U.S. Govt. Print. Off, 1975.

Perrine, M.W. Alcohol involvement in highway crashes: A review of the epidemiological evidence.
In: Schultz, R.D. ed. Clinics in Plastic Surgery 2(1):11-33, 1975b.

Perrine, M.W. Alcohol and highway crashes: Closing the gap between epidemiology and ex-
perimentation. In: Vol. 11, Modern Problems of Pharmacopsychiatry Alcohol. Basel,
Switzerland: Karger, 1976. pp. 22-41.

Perrine, M.W. Analysis Of DUI Processing From Arrest Through Post-Conviction Counter-
measures. Vol. 1..An Evaluation of the California Drunk Driving Countermeasure System.
Technical Report CAL-DMV-RSS-84-89. Sacramento, CA: California Department of
Motor Vehicles, 1984. ,

Perrine, M.W. Varieties of drunken and of drinking drivers: A review, a research program, and a
model. In: Noordzij, P.C., and Roszbach, R., eds. Alcohol, Drugs and Traffic Safety - T86.
Amsterdam: Elsevier, 1987. pp. 105-113.

Perrine, M.W. Alcohol, trauma and traffic safety: research issucs, needs and opportunities.
Contemporary Drug Problems, 15(1):83-93, 1988.

Perrine, M.W.; Waller, J.A.; and Harris, LS. Alcohol and Highway Safety: Behavioral and Medical
Aspects. Technical Report, DOT HS-800-599. Washington, DC: National Highway Traffic
Safety Administration, 1971. :

Pollack, S. Drinking Driver and Traffic Safety Project, Vol I and II. Los Angeles: Public Systems
Research Institute, University of Southern California, 1969.

Pollack, S.; Didenko, O.; McEachern, A.; and Berger, R.M. “Drinking Driver and Traffic Safety
Report.” Final Report Vol. 3. US Department of Transportation. NHTSA Contract No.
FH-11-7099. University of Southern California, Public Systems Research Institute, 1972.

Raymond, A. Drinking-Driving Convictions - A Symptom of Alcoholism? Melbourne: The St.
Vincent’s Summer School of Studies on Alcohol and Drugs, 1971.

Reed, D.S. Reducing the cost of drinking and driving. In: Moore, M.H., and Gerstein, D.R., eds.
Alcohol and Public Policy: Beyond the Shadow of Prohibition. Washington, DC: National
Academy Press, 1981. pp. 336-387

Reis, R.E. The Traffic Safety Effectiveness of Education Programs for First Offense Drunk Drivers.

, Technical Report, DOT Contract HS-6-01414. Washington, DC: National Highway Traf-
fic Safety Administration, 19822.

Reis, R.E. The Traffic Safety Effectiveness of Education Programs for Multiple Offense Drunk
Drivers. Technical Report DOT Contract HS-6-01414. Washington, DC: National High-
way Traffic Safety Administration, 1982b.

Richardson, H.A. Motor Vehicle Traffic Accidents as a Leading Cause of Death in the United States.
Research Notes, Number 85-4. Washington, DC: NHTSA, US Dept of Transportation,
October 1985.

Sadler, D.D., and Perrine, M.W. The Long-Term Traffic Safety impact of a Pilot Alcohol Abuse
Treatment as an Alternative to License Suspensions. Vol. 2: An Evaluation of the California
Drunk Driving Countermeasure System. Technical Report, CAL-DMV-RSS-84-90.
Sacramento, CA: California Department of Vehicles, 1984.

75
76

BACKGROUND PAPERS

Scoles, P., Fine, E.W., and Steer, R.A. Personality characteristics and drinking patterns of high-risk
drivers never apprehended for driving while intoxicated. Journal of Studies on Alcohol
45:411-416, 1984. . a

Simpson, H.M. “The Impaired-Driver Problem vs. the Impaired Problem Driver.” Paper
presented at the Association of Life Insurance Medical Directors of American. Toronto,
Canada, 1977. : . :

Skinner, H.A., and Allen, B.A. Alcohol dependénce syndrome: measurement and validation.
Journal of Abnormal Psychology 91:199-209, 1982. co

Steer, R.A, Fine, E.W., and Scoles, PE. Classification of men arrested for driving while intoxi-
cated, and treatment implications. Journal of Studies on Alcohol 40(31):22-229, 1979.

‘stewart, K, Epstein, L., Gruenewald, P., Laurence, S., and Roth, T. The California First DUT

’ Offender Evaluation Project. Prepared for California Office of Traffic Safety. Walnut
Creek, CA: Pacific Institute of Research and Evaluation, 1987. .
Struckman, D.L. An Analysis of Drinker Diagnosis and Referral. SD: ASAP Analytic Study No. 5 -
1975. Contract No. DOT-HS-045-1-061. Washington, DC: National Highway Traffic
Safety Administration, 1975. :
Sutker, P.B., Brantley, PJ., and Allain, AN. MMPI response patterns and alcohol consumption
in DUl-offenders. Journal of Consulting and Clinical Psychology 48(3):350-355, 1980.
Tashima, H.,and Peck, R. C.An Evaluation of the Specific Deterrent Effects of Alternative Sanctions
for First and Repeat DUI Offenders. Vol 3. An Evaluation of the California Drunk Driving
Countermeasure System. Technical Report CAL-DMV- RSS-86-95. Sacramento, CA:
California Department of Motor Vehicles, 1986.
Vingilis, E.R. Drinking drivers and alcoholics: Are they from the same population? In: Smart, R.G.;
Glaser, F.B.; Israel, Y.; Kalant, H.; Popham, R.; and Schmidt, W. eds. Research Advances
in Alcohol and Drug Problems Vol. 7. New York: Plenum Press, 1983. pp. 299-342
Voas, R.B. ASAP program evaluation methodology and overall program impact. In: ASAP
Evaluation of Operations, 1972. Technical Report, DOT-HS-880-874. Washington, DC:
National Highway Traffic Safety Administration, 1972.
Voas, R.B. Drinking and Driving: Scandinavian Laws, Tough Penalties, and United States Alterna-
tives. Technical Report, DOT Contract DTNH-22-82-P-05079. Washington, DC: Nation-
al Highway Traffic Safety Administration, 1982.
Voas, R.B., and Hause, J.M. Deterring the drinking driver: The Stockton experience. Accident
Analysis and Prevention 19(2):81-90, 1987.
Voas, R.B., Rhodenizer, A.E., and Lynn, C. Evaluation of Charlottesville Checkpoint Operations.
Technical Report, DOT Contract DTNH-22-83-C-05088. Washington, DC: National
Traffic Safety Administration, 1985.
Wells-Parker, E.; Cosby, P.; and Landrum, J.W. Classifying the DUI Offender: A Cluster Analysis
of Arrest Histories. Technical Report. Mississippi: Mississippi State University, 1985.
Wendling, A., and Kolodij, B. An Evaluation of the El Cajon Drinking Driver Countermeasure
Program. Washington, DC: National Highway Traffice Safety Administration, 1977.
Wolfe, A.C. 1973 U.S. National Roadside Breathtesting Survey: Procedures and Results. Interim
Technical Report, DOT-HS-801-241, NTIS No. PB-238-042. Washington, DC: National
Highway Traffic Safety Administration, 1974.
Wolfe, A.C. National Roadside Breathtesting Survey: Procedures and Results. Washington, DC:
Insurance Institute for Highway Safety, 1986.
Zylman, R. Analysis of studies comparing collision- involving drivers and non-involved drivers.
Journal of Safety Research 3:116-128, 1971.
Zylman, R. A critical evaluation of the literature on ‘alcohol involvement’ in highway death.
Accident Analysis and Prevention 6(2), 1974.