Psychological Assessment
1994, Vol.
6, No. 1, 33-40
Copyright 1994 by the American Psychological Association, Inc.
1040-3590/94/$3.00
Personality and Drunk Driving: Identification of DUI Types
Using the Hogan Personality Inventory
Yola Nolan, John A. Johnson, and Aaron
L.
Pincus
Two hundred persons arrested for driving under the influence (DUI), 30 social drinkers, 30 depressed
patients, 30 incarcerated criminals, and 30 alcoholics completed the Hogan Personality Inventory
(HPI) and Court Reporting Network (CRN) interview. A cluster analysis of HPI scores for the DUI
group revealed
5
personality types: Impulsive-Extravert, Normal, Neurotic-Introvert, Neurotic-
Hostile, and Unassertive-Conformist. The types differed predictably on demographic variables,
drinking behavior, and driving records as assessed by the CRN. The Impulsive-Extravert and Nor-
mal types had HPI profiles similar to social drinkers. The Neurotic-Introvert type most resembled
depressed patients, and the Neurotic-Hostile type most resembled incarcerated criminals. Results
clarify previous findings on DUI personality types and establish a basis for tailoring therapeutic
treatments to different types of DUI offenders.
Recent advances in the field of personality structure have
converged, providing a consensus regarding the superordinate
personality dimensions underlying a comprehensive taxonomy
of personality traits
(Digman, 1990; McCrae
&
John, 1992).
The Five-Factor Model
(FFM)
of personality hierarchically or-
ganizes personality traits into five basic dimensions:
I.
Extraver-
sion,
11. Agreeableness, 111. Conscientiousness, IV. Emotional
Stability (vs. Neuroticism), and
V. Intellect or Openness to Ex-
perience. These dimensions broadly assess important individ-
ual differences in interpersonal, motivational, emotional,
atti-
tudinal, and experiential style. This convergence has provided a
working model allowing personality psychologists to proceed
with the conceptual and empirical tasks they have long set for
Yola Nolan, CONCERNS Counseling and Consultation Firm, Du-
Bois, Pennsylvania; John A. Johnson, Department of Psychology, Penn-
sylvania State University, DuBois Campus; Aaron
L.
Pincus, Depart-
ment of Psychology Pennsylvania State University, University Park
Campus.
This article is based on a dissertation written by
Yola Nolan under the
supervision of John A. Johnson in partial fulfillment of the require-
ments for a
PhD from the Professional School of Psychological Studies,
San
Diego, California. Portions of the data reported here were presented
at the 97th Annual Convention of the American Psychological Associa-
tion in New Orleans, Louisiana, August, 1989.
Some of the writing
ofthis article was completed by John A. Johnson
while he was a visiting research fellow at Universitat Bielefeld, Bielefeld,
Germany, supported by a fellowship from the Alexander von
Hum-
boldt-Stiftung. He wishes to express his gratitude to the psychology de-
partment at Bielefeld and the von Humboldt-Stiftung for their support
during that time.
We also thank James Fugate, MD, Medical Director,
Clearfield-
Jefferson Community Mental Health Center; Rita Kartavich, MSA, As-
sistant Administrator of Outreach, DuBois Regional Medical Center;
and Robert Dombrosky, Warden, Clearfield County Jail, for their assis-
tance in our data-gathering. We also thank Michael Gerg for assisting
us in the data analysis and all persons who volunteered to participate in
this study.
Correspondence concerning this article should be addressed to
Yola
Nolan, CONCERNS, 90 Beaver Drive, DuBois, Pennsylvania 15801.
themselves (Wiggins, 1992). Thus, there is currently a large
body of research activity applying the FFM to diverse and im-
portant realms of inquiry including psychotherapy (Miller,
199
I), psychopathology (e.g., Widiger
&
Trull, 1992; Wiggins
&
Pincus, 1989), clinical assessment (e.g., Costa
&
McCrae,
1992a, 1992b; Muten, 1991), health psychology (Smith
&
Wil-
liams,
1992), job performance (Barrick
&
Mount, 1991), and
close relationships (Buss, 1992).
Previous work investigating personality characteristics of
driving-under-the-influence (DUI) offenders and personality
variables associated with drunk-driving accidents has suggested
that the population of DUI offenders contains distinct person-
ality subtypes (Donovan
&
Marlatt, 1982) and that personality
variables play a greater role in drunk-driving accidents than
clinical alcoholism (Clay, 1972; Donovan. Marlatt,
&
Salzberg,
1983; Donovan, Queissler. Salzberg,
&
Umlauf, 1985; Scoles,
Fine,
&
Steer, 1984; Snowdon, 1984). This work was conducted
and interpreted before the emergence and application of the
FFM
of personality. In this article we investigate the personality
types found in a sample of DUI offenders from an FFM frame-
work using a well-validated measure of the FFM, the Hogan
Personality Inventory (HPI; R. Hogan, 1986). The HPI is one
of the major instruments available to assess the FFM
(Briggs,
1992; Trapnell, Wiggins,
&
Broughton, 1992; Wiggins
&
Pin-
cus, 1992, in press; Wiggins
&
Trapnell, in press). Research us-
ing the HPI has emphasized predictive validity, particularly in
occupational settings
(e.g., J. Hogan
&
Hogan, 1986), and the
incorporation of Hogan Assessment Systems has furthered de-
velopment and market applications of the HPI in government
and industry. The theoretically grounded HPI, with its excellent
track record across diverse, applied settings, may prove to be a
useful instrument for identifying personality types at risk for
DUI and for understanding, preventing, and correcting
drunk-
driving behaviors.
Reconceptualizing Donovan and Marlatt's
DUI
Types in
Terms of the Five Factor Model
Our study has a number of empirical goals and hypotheses.
First, we attempted to replicate Donovan and Marlatt's (1982)
34
Y. NOLAN, J. JOHNSON, A. PINCUS
DUI personality subtypes with a better validated measure of
personality. They identified their personality subtypes on the ba-
sis of cluster analysis of a questionnaire containing demo-
graphic variables, questions assessing drinking-related behav-
iors, and 17 short scales measuring driving attitudes and per-
sonality traits linked to accident risk in past research. We
combined the clinical experience of
Y.
Nolan in treating DUI
clients with an FFM interpretation of Donovan and Marlatt's
(1 982) empirically derived DUI clusters to generate our hypoth-
eses regarding personality subtypes in our sample. Consistent
with previous investigations, we expected to find significant
differences in both personality styles and important demo-
graphic and drunk-driving variables across DUI personality
types.
The descriptions for these hypothesized types were informed
by Johnson's (in press; Johnson
&
Ostendorf, 1993) analysis of
FFM-based types. Following a suggestion made by Hofstee and
De Raad
(1992), Johnson labeled profile patterns according to
pairs of scales on which individuals receive relatively high or
low scores. For example, an individual scoring high on Extra-
version but low on Agreeableness is labeled a Domineering Type
or I
+
I1
-
Type.
Hypothesized
FFM
High-Point Codes for
DUI
Types
Impulsive-Extravert (I
+
111
-)
Donovan and Marlatt (1982) identified a relatively assertive,
uninhibited, externally oriented DUI group at moderate risk
relative to other clusters. They exhibited good social adjustment
and relatively light drinking. We hypothesized that from an
FFM perspective, such a
DUI group would present as extra-
verted and impulsive.
Neurotic-Introvert (I
-
IV
-)
Donovan and Marlatt (1 982) found a relatively unassertive,
depressed, inhibited, irritable DUI cluster with little sense of
personal control in their lives. This cluster was at relatively
greater risk for DUI. From an
FFM
perspective, we predicted
such a group would present as neurotic and introverted.
Neurotic-Hostile (11
-
IV
-)
Donovan and Marlatt (1982) identified an aggressive, poorly
adjusted, sensation-seeking, heavy drinking DUI cluster also at
relatively high risk for DUI. From an
FFM perspective, we pre-
dicted such a group would present as neurotic and hostile.
Unassertive- Conformist (I
-
V
-)
Donovan and Marlatt (1982) found a DUI cluster that also
drank heavily but otherwise reported adequate adjustment and
relatively few accidents. They considered this cluster to be sim-
ilar to but less deviant than the Neurotic-Hostile cluster; how-
ever, data indicated that this group appeared to be preoccupied
with maintaining a sense of personal control; was significantly
less hostile, resentful, competitive, and irritable; and more in-
hibited than the former cluster. The individuals in this cluster
seemed unassertive and lacking in drive and energy, preferring
to maintain a sense of control. From an
FFM perspective, we
hypothesized that such a group may be better described as un-
assertive (rather than neurotic) and conformist (rather than hos-
tile) in using alcohol to reduce tension.
Normal
Donovan and Marlatt (1982) also found a cluster with no dis-
cernibly abnormal characteristics. Compared with other clus-
ters, this group had higher social status, a good driving record,
and relatively low levels of drinking. We had no a priori hypoth-
esis regarding the FFM description of such a cluster, although
we predicted the members would not present as significantly
maladjusted.
Socioanalytic theory (R. Hogan, 1983) regards psychological
disorders as forms of self-presentation that are continuous with
forms of self-presentation in the normal population. However,
Lorr and
Strack (1993) pointed out that subgroups based on
FFM dimensions may differ across different types of popula-
tions. We therefore selected four comparison groups, which we
predicted would demonstrate differential similarity to derived
DUI clusters, to more accurately interpret the characteristics
of cluster members. Specifically, we predicted that
Impulsive-
Extravert DUIs and Normal DUIs would be most similar to a
group of social drinkers, Neurotic-Introvert DUIs would be
most similar to a group of depressed patients, and
Neurotic-
Hostile DUIs would be most similar to a group of incarcerated
criminals. We also included a comparison group of alcoholics.
Method
Overview
We cluster analyzed HPI scores for a group of first time DUI offend-
ers. We used three different clustering methods and then compared the
similarity of cluster solutions. Following cluster identification, we com-
pared these offender types on variables not included in the original clus-
ter analysis. These included age, income, drinking attitudes, behavioral
impairment from drinking, and quantity and frequency of alcohol con-
sumption. We then compared HPI profiles from our DUI clusters with
non-DUI comparison groups including social drinkers, clinically de-
pressed patients, incarcerated criminals, and clinical alcoholics.
Subjects
DUZs.
This sample consisted of 200 first time DUI offenders (1 84
males,
16
females; average age 32.47 years) arrested for driving a motor
vehicle while at the "per se level" of intoxication
(.lo blood alcohol
level in Pennsylvania, measured by certified breath or blood testing).'
Subjects were referred to Y. Nolan for evaluation by the county court
system. Two subjects were African-American, the remainder were
White. In this sample, 22.5% were single. 38.590 were married, 33.0%
were separated, 3.590 were divorced, and 2.590 were widowed.
With regard to occupation, 34.7% were unemployed; 27.2% were un-
skilled or semiskilled workers
(e.g., custodian); 12.7% were skilled work-
ers
(e.g., mechanic); 9.8% worked in clerical, sales, or technician posi-
tions
(e.g., bookkeeper); 12.1% worked as midlevel professionals (e.g.,
manager); and 3.5% were professionals (e.g., lawyer). Years of formal
education ranged from a minimum of 5 years to a maximum of 19
years, with a mean of
1
1.7
years.
'
All subjects were volunteers who participated without compensa-
tion. Individuals arrested for DUI were excluded from comparison
groups.
DUI PERSONALITY TYPES
3
5
Social drinkers.
This comparison group consisted of 30 social
drinkers (6 males, 24 females; average age 33.0 years) who responded to
an advertisement to participate in the study. These persons reported
ingesting no more than two drinks per day, had never experienced life
problems through the use of alcohol, and had no alcohol related health
problems. One subject was Asian, the remainder were White. In this
sample, 76.7% were married, 20.0% were single, and 3.3% were di-
vorced. With regard to occupation, 26.7% were unemployed; 3.3% were
unskilled or semiskilled workers; 10.0% were skilled workers; 16.7%
worked in clerical, sales, or technician positions; 30.0% worked as
mid-
level professionals; and 13.3% were professionals. Years of formal edu-
cation ranged from a minimum of 10 years to a maximum of 20 years,
with a mean of 14.3 years.
Depressed patients.
The second comparison group consisted of 30
patients from a rural Pennsylvania community mental health center (6
male, 24 female; average age 36.9 years) who were diagnosed as de-
pressed upon intake with the Minnesota Multiphasic Personality Inven-
tory (MMPI) and who were still being treated at the time for depression.
Subjects were recruited by the director of the center. All subjects were
White. In this sample, 33.3% were married, 26.7% were single, 10.0%
were separated, 26.7% were divorced, and 3.3% were widowed. With
regard to occupation, 60.0% were unemployed; 0.0% were
unskilled or
semiskilled workers; 3.3% were skilled workers; 13.3% worked in cleri-
cal, sales, or technician positions; 23.3% worked as
midlevel profession-
als; and 0.0% were professionals. Years of formal education ranged from
a minimum of 6 years to a maximum of 19 years, with a mean of 12.8
years.
Incarcerated criminals.
The third comparison group consisted of
30 criminals (26 male, 4 female; average age 28.7 years) serving jail
terms in a Pennsylvania county prison for reasons other than driving
under the influence. Subjects were recruited by the prison warden. Two
prisoners described themselves as Hispanic, the remainder were White.
In this sample, 26.7% were married, 50.0% were single, 3.3% were sep-
arated, 16.7% were divorced, and 3.3% were widowed. With regard to
occupation, 83.3% were unemployed; 3.3% were
unskilled or semi-
skilled workers; 0.0% were
skilled workers; 10.0% worked in clerical,
sales, or technician positions; 3.3% worked as
midlevel professionals;
and 0.0% were professionals. Years of formal education ranged from a
minimum of 9 years to a maximum of 16 years, with a mean of 11.8
years.
Alcoholics.
The final comparison group consisted of 30 alcoholics
(23 male, 7 female; average age 29.3 years) recruited from drug and
alcohol agencies, Alcoholics Anonymous, and hospital and rehabilita-
tion centers in rural northwest Pennsylvania. All subjects were White.
In this sample, 26.7% were married, 33.3% were single, 20.0% were sep-
arated, 16.7% were divorced, and 3.3% were widowed. With regard to
occupation, 76.7% were unemployed; 13.3% were
unskilled or semi-
skilled workers; 6.7% were skilled workers; 0.0% worked in clerical,
sales, or technician positions; 3.3% worked as
midlevel professionals;
and 3.3% were professionals. Years of formal education ranged from a
minimum of 7 years to a maximum of 16 years, with a mean of 10.9
years.
Measures
After signing an informed consent form, all subjects completed the
HPI (R. Hogan, 1986). The HPI is a 310 item true-false self-report
inventory assessing a six-factor variant of the
FFlM of personality
(Briggs, 1992; McCrae
&
John, 1992). The HPI also contains a short
validity scale to check for careless responding. Alpha reliabilities for
scales range from .76 to
.89; 4-week test-retest reliabilities range from
.74 to
.99. The scales have been validated in samples (total
N>
2,000) of
adult men and women employed in a variety of occupations (R. Hogan,
1986).
Y. Nolan, who is trained in administration and scoring, administered
the Court Reporting Network Evaluation (CRN; Scoles &Cook, 1986).
The CRN evaluation consists of 87 questions that the interviewee an-
swers in a 40-minute session. The CRN contains several items concern-
ing such basic demographic information as age, sex, and income. The
CRN also contains three scales: the Mortimer-Filkins (MF) index,
which identifies factors related to problem drinking; an impairment
index (IMP), which measures alcohol-related physical and psychologi-
cal impairment; and a quantity-frequency index of alcohol consump-
tion (QF).
The MF is an empirically constructed scale, organized around areas
of
marital and family problems, recent stress, financial difficulties, per-
sonal adjustment problems, previous arrests, driving history, and alco-
hol dependence. Scoles and Cook (1 986) reported a split-half reliability
coefficient of .97 and validity coefficient (point-biserial correlation be-
tween an alcoholic and control group) of .92 for the
Mf
scale.
The IMP consists of 12 items assessing behavioral problems caused
by excessive alcohol use (blackouts, shakes, missing meals or work, etc.).
The QF measures the amount and frequency of beer. wine: and liquor
consumption within a 30-day period. A separate QF for each beverage
is calculated by multiplying the self-reported frequency of consumption
(in a month) by the self-reported quantity consumed (on an average
occasion) by the estimated proportion of ethanol content (by volume)
in the beverage. The total QF index is the sum
ofthe individual beverage
indices. Both the IMP and the QF have been extensively field tested and
validated by the National Institute on Alcohol Abuse and Alcoholism.
The QF index has been criticized for using "five or more drinks" as the
upper bound for assessing quantity consumed. This potentially creates
artificially low ceiling effects in assessment.
Analyses
The program k-means (PKM) routine from the BMDP statistical
package (Dixon et al., 1983) was applied to the standardized scores on
the six primary HPI scales for the 200 DUI offenders. PKM uses a
k-
means clustering technique and establishes a fixed number of homoge-
neous groups of cases using Euclidean distances (Lorr, 1983). PKM be-
gins with all cases assigned to one cluster. The program repeatedly splits
this and subsequent clusters until it produces a requested number of
clusters. Cases are iteratively reassigned to a cluster whose center (mean)
is closest, as measured by Euclidean distance.
Lorr and Suziedelis (1990) have indicated that more than two dozen
arbitrary rules exist for determining how many clusters one should ex-
tract in an analysis; several Monte
Carlo studies have compared the uti-
lity and validity
ofthese rules when used with a variety ofcluster analy-
sis programs
(e.g., Milligan, 198 1). These studies indicated that perfor-
mance of cluster algorithms can vary with data characteristics although
some are relatively more consistent than others (Milligan
&
Cooper,
1985). We selected a k-means algorithm for initial cluster derivation
because this procedure produces "excellent recovery of structures when
the starting seeds are obtained from the group-average method or when
valid a priori information is available" (Lorr, 1983, p. 117). Because
we hypothesized finding FFM-based clusters similar to those previously
identified by Donovan and Marlatt
(1982), the PKM program was in-
structed to identify five clusters. This decision was also influenced by
the suggestion that cluster analyses of measures of personality from both
normal and alcoholic populations generally identify four to seven clus-
ter types (Lorr
&
Strack, 1993; Lorr
&
Suziedelis, 1990).
To assess the stability of the five-cluster solution, we conducted two
additional cluster analyses. First, we used the cluster means from the
PKM solution to define initial seeds for a Quick Cluster analysis (SPSS,
1988) of HPI scores. Quick Cluster begins with one cluster and itera-
tively reassigns cases to the nearest cluster according to squared Euclid-
ean distance until the requested number of clusters is reached.
Finally, Ward's (1963) hierarchical clustering procedure from the
Cluster routine (SPSS, 1988) was applied to the HPI scores. Unlike the
two previous techniques, which subdivide the entire sample into smaller
clusters, this procedure begins with individual cases and iteratively
ag-
3
6
Y
NOLAN,
J.
JOHNSON, A.
PINCUS
Table
1
Hogan Personality Inventory (HPZ) Profiles for Driving-Under-the-Influence Clusters and Comparison Groups
Cluster or erou~
HPI
factors
N
INT
ADJ
PRU
AMB
SOC
LIK
Impulsive-
Extravert
49
48.8
46.7
41.3
49.8
55.7
52.0
Neurotic-
Normals Introvert
45 32
49.0
35.7
50.7
33.4
58.3
44.4
41.4
32.8
43.0
40.7
55.4
42.0
Neurotic-
Hostile
36
45.3
3 1.4
44.2
47.9
48.9
32.4
Unassertive-
Conformist Total DUI
3 8 200
28.6 41.9
46.2
42.6
56.2
49.4
28.4 40.7
42.0 46.6
47.8
46.8
Social
F(4, 195) drinkers
-
30
47.9* 45.2
48.3* 45.1
36.2* 48.8
48.7* 45.8
37.0* 52.7
67.0* 51.0
Depressed
patients
30
40.3
23.2
50.4
34.7
43.6
40.0
Criminals Alcoholics
30
40.9
29.7
37.8
45.1
54.3
41.7
Note.
Bold means indicate smallest difference significant at the .05 level with a Scheffe posttest.
T
scores computed from normative data
for males presented in R. Hogan (1986). INT
=
Intellectance: ADJ
=
Adjustment; PRU
=
Prudence; AMB
=
Ambition; SOC
=
Sociability;
LIK
=
Likability.
*p< ,001.
glomerates them into larger and larger clusters by optimizing an objec-
tive function known as the within-groups sum of squares, or ESS (Lorr
&
Strack, 1993). Merger of cases into clusters occurs on the basis of
a minimum increase in ESS. The optimal number of clusters can be
determined from this method by noting a marked discontinuity in the
proximity coefficients between levels of clustering.
The groups identified through cluster analysis were then compared
with an analysis of variance
(ANOVA) with respect to demographic
variables and the MF, IMP, and QF indices.
The similarities among DUI clusters and comparison group HPI pro-
files were examined through Mahalonobis
@ values (Rao, 1952). Ma-
halanobis
D2
values provided a statistical test for significant multivariate
differences between DUI clusters and comparison group profiles. We
then examined relative similarities between the DUI clusters and com-
parison group profiles through Cronbach's
@,
a descriptive measure
better suited for comparing profile similarity values (Cronbach
&
Gleser, 1953; Wiggins, 1973). After we identified DUI and comparison
group similarities, we compared groups through a discriminant func-
tion analysis program (Nie, Hull, Jenkins, Steinbrenner,
&
Bent, 1975).
The intent was to determine whether cases from each DUI cluster would
be misidentified most frequently as belonging to a particular compari-
son group and whether cases from the comparison groups would be
misidentified most frequently as belonging to a particular DUI cluster.
Such differentially linked "misidentifications" would provide addi-
tional indications of the degree of personality similarity between DUI
subtypes and particular comparison groups.
Results
Cluster Analyses
The PKM clustering procedure partitioned the
DUI
sample
into five groups with distinctive profiles on the HPI.
A
multivar-
iate analysis of variance using cluster membership as the group-
ing variable and the six HPI scores as dependent variables pro-
duced an
F(6, 190)
value of
3604.94,
p
<
.00001.
Univariate
ANOVA showed significant differences among the clusters on
every HPI scale (see Table
1).
Table
2
Court Reporting Network Evaluation (CRN) Profiles for Driving- Under-the-Influence Clusters and Comparison Groups
Cluster or group
CRN Impulsive- Neurotic- Neurotic- Unassertive- Total Social Depressed
variables
Extravert Normals
Introvert Hostile Conformist DUI F(4, 195) drinkers patients Criminals Alcoholics
N 49 45 32 36
38
200
-
30 30
30 30
Age 3 1 .O 34.0
29.3
28.6
38.9
32.5 4.39* 33.0 36.9 28.7 29.3
13.2 12.8 9.6 11.6 12.8 12.6 11.9 12.3 8.0 10.6
Inc 14.041
15,022
11,594
7,361
14,790 12,810 3.51* 20,233 19,448 7.607 10,567
10,077 12,754 10,245 6,779
1 1,376 10,807 14,229 17,659 12,102 11,982
MF 48.2
48.8
71.1
71.7
63.7 59.2 5.33* 16.8 38.1 83.9 116.4
30.1 32.4 36.3 33.6 27.9 33.3 14.2 27.2 44.3 33.3
IMP^^
3.00 2.96 4.84 5.31 3.42 3.78 3.51* 0.64 0.43 6.67 14.07
2.33 3.39 4.03 5.30 2.94 3.72 1.62 1.37 5.88 8.30
QF
.72 .7
1
.72 1.04 .73 .78 .40 .I5 .06 2.96 4.39
1.23 1.56 1.17 1.60 1.30 1.38 .28 .19 6.02 3.49
Note.
Standard deviations are reported below mean scores. Bold means indicate smallest difference significant at the .05 level with a Scheffe post-
test. Inc
=
Income; MF
=
Mortimer-Filkins index; IMP
=
Impairment index; QF
=
QuantityIFrequency index.
a
Cut-offs for problem drinking (Scoles
&
Cook, 1986): MF
=
25.0, IMP
=
5.90, QF
=
2.00. Overall ANOVA significant but no two groups
different by
Scheffe posttest.
*p
<
.01.
DUI PERSONALITY TYPES
Table 3
Indices
of
Profile
Similarity
for
Driving-Under-the-Injluence
(DUI) and Rekrence Groups
Similarity to reference groups
DUI clusters Social drinkers Depressed patients Criminals Alcoholics
Impulsive-Extravert 43.74
i
.92
Normal 100.86
2.84
Neurotic-Introvert 194.94
7.70
Neurotic-Hostile
18 1.50
6.74
Unassertive-Conformist 20 1.84
8.90
Note.
Top number is Cronbach's D2; second number is Mahalanobis D2. The smaller the value ofD2, the
more similar the groups. All Mahalanobis values are significant at thep
<
.OO
1
level.
When the DUI cases were reclustered with the Quick Cluster
method, 80% of the PKM Cluster 1 cases reemerged in the same
cluster, 87% from PKM Cluster 2 remained together. 9 1% from
PKM Cluster 3 remained together, 83% from PKM Cluster 4
reemerged together, and 89% of the
PKM Cluster 5 cases reap-
peared in the same cluster. When subjected to the significantly
different Ward's procedure, the following percentages of cases
from PKM clusters one through five reemerged in the same
cluster:
65%, 96%. 31%, 72%, and 92%. Only the third PKM
cluster appeared somewhat unstable: Ward's procedure
grouped a significant number (53%) of cases from this cluster
with Ward's Cluster 5. The change in proximity coefficients be-
tween 3-cluster and 7-cluster levels of the grouping hierarchy
indicated that either four or five clusters be considered optimal.
Cluster membership was judged to be stable across methods,
and the remaining analyses were based on the five clusters iden-
tified in the PKM analysis.
Univariate ANOVA across CRN variables revealed that DUI
clusters differed significantly in terms of age, income, MF
scores, and IMP, but not in terms of the QF Index. Means. ex-
pressed as
T
scores based on normative data for the HPI (R.
Hogan,
1986), are shown for the DUI clusters and four compar-
ison groups in Table
1
;
CRN information for all groups is shown
in Table 2.
HPI ProJile Comparisons
Table 3 shows both the Mahalonobis
D2
and Cronbach
D~
values of profile similarity. Mahalanobis
d
values showed that
all DUI cluster profiles differed significantly from the profiles of
the four comparison groups. Thus, none of the DUI clusters was
!
found to "belong to" a comparison population in a statistical
sense. The Cronbach
D2
values clearly demonstrate, however,
I
strong relative differences between each DUI cluster and the
four non-DUI groups.
Discriminant Analvsis
Only the first discriminant function showed an eigenvalue
greater than unity (1.4). This function accounted for 46.5% of
the variance and classified 62.5% of the 320 cases correctly. Ta-
ble
4
shows the hit and miss rates for the discriminant analysis.
The primary misclassifications aided in interpretation of the
DUI clusters. For example, cases from the DUI Cluster 1 were
misclassified most often as social drinkers and social drinkers
were misclassified most often as Cluster
1
DUIs, indicating per-
sonality similarity between these two groups.
Discussion
The distinctive pattern of low and high scores on the HPI and
CRN and on the relative similarity to the four comparison
groups supported the existence of five distinct DUI types hy-
pothesized at the outset of the study. A summary of DUI clus-
ters' distinctive patterns follows.
Cluster
1
flmuulsive-Extruvertj
The HPI profile of the first DUI cluster is most similar to
the profile of social drinkers. Both groups showed relatively flat
profiles hovering about the
T
=
50 level with a slight elevation
on the Sociability (If) scale. Cluster
1
differed from the social
drinker group only in a somewhat lower Prudence scale
(111-),
indicating a greater degree of impulsivity and self-indulgence in
the DUIs (R. Hogan, 1986). Cluster 1 had the second lowest
IMP score
(3.00), and the lowest MF score (48.2) of the five DUI
clusters; nonetheless, the CRN manual states that MF scores
greater
than 38 suggest a pathological drinking pattern. These
considerations led to the label
Impulsive-Extrmlert
for this DUI
cluster.
Cluster
2
(Normal)
This cluster resembled the social drinkers more than any
other comparison group, although the similarity is less than that
found between the Impulsive-Extravert DUIs and social drink-
ers. Normal DUIs are characterized by average lntellectance
and Adjustment, peaks on Prudence and Likability, and slightly
depressed scores on Ambition and Sociability. The higher Pru-
dence score and lower Ambition and Sociability scores distin-
guish Cluster
2
from the Impulsive-Extravert DUIs. People
with profiles similar to Cluster 2s are described by others with
