Personality characteristics of sex offenders : a ... · iii . offenses 'that most often included...
Transcript of Personality characteristics of sex offenders : a ... · iii . offenses 'that most often included...
PERSONALITY CHARACTERISTICS OF SEX OFFENDERS:
A COMPARISON OF THE MMPI AND MCMI
Shelly C. OfConnor
B.A. (Hons), University of Victoria, 1986
THESIS SUBMITTED IN PARTIAL FULFILLMENT OF
THE REQUIREMENTS FOR THE DEGREE OF
MASTER OF ARTS
, Department
of
Psychology
@ Shelly C. OIConnor 1990
SIMON FRASER UNIVERSITY
November 1990
All rights resewed. This work may not be reproduced in whole or in part, by photocopy
or other means, without permission of the author.
APPROVAL
Name : Shelly Charlene OIConnor
Degree: Master of Arts
Title of Thesis: Personality characteristics of Sex Offenders:
A comparison of the MMPI and MCMI
Examining Committee:
Chairperson: Dr. Marilyn Bowman
Dr. Richard J. Freeman Senior Supervisor
Dr. DWid N. Cox
Dr. Carson W. smiley External Examiner Director of Psychosocial Services Regional Psychiatric Centre Correctional Services of Canada
Date Approved: -
PARTIAL COPYRIGHT LICENSE
I hereby grant t o Simon Fraser U n i v e r s i t y the r i g h t t o lend
my thes i s , p r o j e c t o r extended essay ( the t i t l e of which i s shown below)
t o users o f the Simon Fraser U n i v e r s i t y L ib ra ry , and t o make p a r t i a l o r
s i n g l e copies on l y f o r such users o r i n response t o a request from the
l i b r a r y o f any o the r u n i v e r s i t y , o r o ther educat ional i n s t i t u t i o n , on
i t s own beha l f o r f o r one o f i t s users. I f u r t h e r agree t h a t permission
f o r m u l t i p l e copying o f t h i s work f o r scho la r l y purposes may be granted
by me o r the Dean o f Graduate Stydies. I t i s understood t h a t copying
o r p u b l i c a t i o n o f t h i s work f o r f i n a n c i a l gain s h a l l no t be al lowed
w i thou t my w r i t t e n permission.
T i t l e o f Thesis/Project/Extended Essay
Personality Characteristics of Sex Offenders: -.
A Coqarison of the MMPI and MCMI
Author:
( s i g$ature)
: Shelly Charlene OIConnor
(name)
(date)
Abstract
This study examined the relative utility of the
Minnesota Multiphasic Personality Inventory (MMPI) and the
Millon Clinical Multiaxial Inventory (MCMI) in
differentiating adult males who had committed sexual
offenses. To date, the MMPI has been largely unsuccessful
at differentiating types of sex offenders. A partial
explanation for inconsistent results is the relative
infrequency in the sex offender population of those clinical
syndromes xis I of DSM-111-R) best captured by the MMPI.
It was hypothesized that sex offenders are more likely to
harbour personality disorders (Axik 11), which the MCMI is
specifically designed to assess.
A sample of 127 adult males who had committed sexual
offenses was studied using personality factors (derived from
the MMPI and MCMI separately) and offender characteristics
to predict offense type. Results showed that offender
variables yielded three meaningful solutions when submitted
to cluster analysis. These included an Unskilled Labour
group, an Unemployed group who, and a Skilled Labour group.
Offense/victim variables also yielded three meaningful
cluster solutions. These were Intrafamilial offenses that
typically included one intrafamilial victim who was ten
years old or younger, 20 or more offense repeats of a
relatively serious nature (i.e. the offense included oral
sex; digital, vaginal, and/or anal penetration); ~mpulsive
iii
offenses 'that most often included one extrafamilial victim
of eleven years old or older, a one-time occurrence, and of
a relatively less serious nature (e.g., exhibitionism,
obscene phone calls, or fondling and/or masturbation);
Chronic offenses that included two or more victims who were
ten years old or younger, between two and 19 offense repeats
that were of a relatively serious nature and tended to be
both intrafamilial and extrafamilial.
The MMPI and MCMI each produced four meaningful
personality factors. These personality factors and the
offender clusters were used to predict offense type. The
MCMI demonstrated superior predictive validity in that it
discriminated the Chronic from the Intrafamilial offenses,
as well as the average of these two offense groups from the
Impulsive offenses. The MMPI only discriminated the Chronic >
from the ~ntrafamilial offenses. Implications for treatment
and recidivism were discussed within the context of
personality characteristics.
Acknowledgements
I would like to thank Richard Freeman for his support
and encouragement. Without his gentle nudging, I may still
be working on this. Also, I would like to thank Louis
Sutker for his generous sharing of data and relevant
literature. Thanks go out as well to Angela, Doug, Karen,
and Sue, for their feedback, which made all the difference
in @@Defense Postmortemst1. A special thank you to Michael
for his guidance, patience, and support for the duration of
this project, and most importantly for his Itway of being in
the worldl1. Additional people who were instrumental in
providing much appreciated help were David Cox, Carson
Smiley, and Lorie Tarcea. Finally, I would like to thank my
mother, Gail, for being there in many ways for me throughout I
the course of my graduate career, and my father, Thomas, who
although isnlt quite sure exactly what it is I am doing,
thinks it's wonderful anyway.
TABLE! OF CONTENTS
Approval ......................................... ii Abstract ........................................ iii Acknowledgements ................................. v List of Tables ................................... vii I . Introduction ............................... 1
MMPI Research with Sex Offenders ........... 3 Sample Selection ........................... 5 Definition of Groups ....................... 6 ~efinition of MMPI Protocol Validity ....... 13 Statistical Analysis of MMPI Profiles ...... 15 MMPI Scores Reported ....................... 16 MMPI Results in Sex Offender Studies ....... 17 The Millon Clinical Multiaxial Inventory ... 22 Purpose of Fresent Study ................... 24
I1 . Method ..................................... 25 Subjects ................................... 25 Procedure .................................. 25
I11 . Results .................................... 28 IV . Discussion ................................. 58
................................. V . References 74
Table
LIST OF TABLES
Page
1. Different Groups Used in MMPI ..................... Research with Sex Offenders 9
........... 2. MMPI and MCMI Means for Total Sample 31
........... 3. Order Independent MMPI 2-Point Codes 31
4. Frequency and Percentage of Subjects ......... in Total Sample with MMPJ T-scores >69 32
5. Order Independent MCMI 2-Point Codes ............ and Highest single Scale Elevations 32
6. Frequency and Percentage of subjects ........ in Total Sample with MCMI BR Scores >75 34
7 . Contingency Coefficients Among Offender Variables ............................. 35
8. Contingency Coefficients Among .............................. Offense Variables 38
9. Contingency coefficients Among Offense and Offender Variables ................. 38
? ..................... 10. MMPI Rotated Factor Matrix 40
..................... 11. MCMI Rotated Factor Matrix 42
12. Pearson Correlation Matrix of MMPI and MCMI Factors ............................... 42
13. Set Correlation Analysis of Offender and Personality Variables Related to Coded Set of Offense ...................................... Variables 53
14. Set Correlation Analysis of Intrafamilial Group Versus .................................. Chronic Group 54
15. Set correlation ~nalysis of the ~mpulsive Group Versus the Average of the ~ntrafamilial ............................. and Chronic Groups 55
16. Summary of Set Correlation ...................................... Analysis 56
CHAPTER I
INTRODUCTION
This study investigated the personality characteristics
of various subgroups of sex offenders. More specifically,
Minnesota ~ultiphasic Personality Inventory (MMPI) and
Millon clinical ~ultiaxial Inventory (MCMI) profiles were
compared to determine which instrument better classified sex
offenders into subgroups differing in offense type, victim
characteristics, and offender variables. The section to
follow describes the prevalence and problem of sexual
offending, and reviews the literature on MMPI research with
sex offenders. The section will conclude by introducing the
MCMI, and proposing a rationale for including this
instrument in the present study.
The term sexual offense refers to a wide range of
illegal sexual behaviors, including rape, child molestation,
incest, and indecent exposure. In British Columbia the
number of adults charged with sex offenses has increased
dramatically from 495 in 1980 to 1003 in 1985. These 1003
individuals were charged with 3,669 sex offenses. Thus, an
average of about ten sex offenses were committed every day
(LaTorre, 1987). Given this trend, the numbers of persons
charged with sex offenses likely is even higher at present.
LaTorre (1987) suggests several factors that may have
contributed to this increase, including better reporting
procedures, more aggressive prosecution, and amended
definitions of sexual offenses under the Criminal Code
introduced in 1985. LaTorre (1987) acknowledges, however,
that the increase in charges must partly be attributed to a
real increase in the incidence of sex offenses.
~ccording to an estimate put forth in the Badgley
Report prepared by the federal government in 1984, one in
ten boys and one in four girls are sexually assaulted as
children. Clearly, sex offenses constitute a serious social
problem, and the body of research devoted to this topic is
expanding at a rapid rate.
One issue that has received considerable attention
concerns assessing the personality characteristics of sex
offenders. Research investigating the personality
characteristics of sex offenders typically assesses
personality using the MMPI. It is therefore important to
determine the utility of the MMPI in differentiating sex
offenders from offenders convicted of other non-sex crimes
and in differentiating various subgroups of sex offenders.
Large numbers of persons charged with sexual offense(s)
are referred by the courts for a psychiatric predisposition
assessment each year. In Canada, predisposition assessments
address issues such as fitness to stand trial, treatability
and dangerousness of the offender. These assessments
typically include psychometric and clinical appraisal of the
offender, and aid the court in determining disposition.
Again, the instrument most often employed in psychometric
assessments of sex offenders is the MMPI, which suggests a
second reason for evaluating the MMPI in the context of sex
offenders. The following is a review of the literature
relating the MMPI to sex offenders. Hereafter, WMPI
researchgg refers specifically to MMPI research with sex
offenders.
MMPI research with sex offenders
Early attempts to differentiate subgroups of sex
offenders, or sex offenders from non-sex offenders using
MMPI profiles, employed empirically based subscales such as
the Sexual Deviation scale (Marsh, Hilliard, & Leichti,
1955), the Pedophilic scale (Toobert, Bartelme, & James,
1959), and the Aggravated Sex scale (Panton, 1970).
However, these scales are typified by poor criterion related
validity, and as might be expected from sample specific
measures, rarely cross-validate.
For example, Peek and Storm (1956) and Wattron (1958)
failed to validate the Sexual ~eviation scale (Marsh et al.,
1955), prompting Wattron (1958) to conclude that the
I1instrument is of no practical value in discriminating
between sex offenders and other type felons in correctional
settings." (p.16). The Sexual Deviation scale appears to be
more a measure of gross maladjustment (Wattron, 1958) and
general abnormality (Peek & Storm, 1956) than of deviant
sexual trends.
Butcher and Tellegen (1978) discuss several problems
with the development and use of MMPI subscales. They
conclude that subscales often Itprove to be largely redundant
alternative versions of existing scales, although sometimes
of poorer qua1ity.I' (p.622). Thus, more recent research
relating the MMPI to sex offenders has focused primarily on
the ten standard clinical and three validity MMPI scales.
Unfortunately, this shift from subscales to standard scales
has not produced a clearer picture of the personality
characteristics of sex offenders. Indeed, the more recent
research has produced an inconsistent and ultimately
confusing welter of results. In the sections to follow I
argue that much of this confusion arises from poor design
and methodological variations across studies. The following
issues will be discussed:
1. Samples selection
2. Definition of groups
3. Definition of MMPI protocol validity
4. Statistical analyses of MMPI profiles
5 . MMPI scores reported.
Finally, MMPI results will be discussed in the light of
these methodological problems, and the Millon Clinical
Multiaxial Inventory will be introduced.
1. Sam~le selection
The samples used in sex offender studies have varied on
several dimensions beyond type of offense. For example,
some subjects were incarcereted in prison (Carroll & Fuller,
1971; Panton, 1978; Panton, 1979), whereas others were
remanded to a psychiatric unit for pretrial assessment
(Armentrout & Hauer, 1978; Quinsey, Arnold, & Preusse,
1980). Moreover, these settings vary as to whether they are
maximum security (Quinsey et al., 1980) or medium security
(Carroll & Fuller, 1971).
Subjects also have been selected from a psychiatric
hospital ward for the criminally insane (Anderson, Kunce, &
Rich, 1979), a state hospital in-patient unit (Hall, Maiuro,
Vitaliano, & Proctor, 1986; Hall, 1989), and an out-patient
family treatment program (Scott & Stone, 1986). Finally,
subjects vary according to stage of proceedings; some have
been convicted for their offense (Carroll & Fuller, 1971;
Groff C Hubble, 1978; and Scott & Stone, 1986), some were
referred by the courts for a presentencing evaluation after
conviction (Erickson, Luxenberg, Walbek, & Seely, 1987;
McCreary, 1975a, McCreary, 1975b), and others were referred
to various settings for pretrial assessment (Armentrout &
Hauer, 1978; Quinsey et al., 1980; Rader, 1977). Obviously,
these samples would be expected to differ on a wide variety
of characteristics besides type of sex offense, each of
which could affect MMPI profiles. Thus, one reason why
MMPI-sex offender results may be so inconsistent is because
sex offense and other sample characteristics are confounded
across studies.
2. Definition of srouw
Itsex offensen is an umbrella term referring to a wide
range of behaviors. Consequently, sex offenders represent a
heterogeneous group that varies in offender, offense, and
victim characteristics. Given the enormity of this
variation, any single study addressing offender personality
characteristics would be hard pressed to obtain
representative samples, and once obtained, to find enough
individuals in each combination of offender, offense, and
victim characteristics to reliably differentiate them.
This has had two general effects on studies using the
MMPI to differentiate or classify sex offenders. The first,
and perhaps most problematic of these, is that groups are
categorized along a single dimension with the remaining
dimensions left uncontrolled and free to undermine the
internal validity of the results. The second is that many
studies use a small number of narrowly defined groups. This
second problem is less an issue at the level of a single
study, but hinders interpretation when the narrowly defined
groups vary across studies. As discussed below, such
across-study variation in offender characteristics is the
rule rather than the exception in MMPI research.
The notion that sex offenders make up a heterogeneous
group is not new, and has been proposed by several
researchers (e.g., Armentrout & Hauer, 1980; Hall, 1989;
Rader, 1977). Nevzrtheless, studies (primarily earlier
studies) have combined sex offenders that have committed a
variety of crimes (e.g., rape, incest, indecent exposure)
into one group, and compared it to groups of non-sex
offenders (Carroll & Fuller, 1971; Hartmann, 1967; Swenson &
Grimes, 1958). These early studies yielded few significant
findings. Also, MMPI profiles of heterogeneous groups of
sex offenders have not been found to differ significantly
from profiles of other criminal groups (e.g. Panton, 1978).
Thus, studies that have conbined various types of sex
offenders into one group, and attempted to differentiate
them from groups of non-sex offenders on the basis of MMPI ,
profiles, have, for the most part, been unsuccessful.
More recent studies employing the MMPI to differentiate
types of sex offenders have used more narrowly defined
groups and attempted to better control for competing
differences among groups. However, this refinement has lead
to a new problem, namely that virtually no two studies use
the same groups for comparison. Rather than replicating and
expanding previous studies, each study represents a radical
departure from the studies before it, in terms of group
definition. Table 1 helps to illustrate the numerous
different between the groups selected for MMPI research
across studies.
The major difficulty that arises from the using
numerous group types across studies is that direct
comparison of results is often confusing and sometimes
impossible. his is especially true when the issue of
heterogeneous sample selection is also considered.
MMPI studies often have selectively chosen specific
dimensions of offender, offense, or victim characteristics
for investigation. Typically, groups are defined on the
basis of one of these, followed by a comparison of
personality characteristics derived from each group's MMPI
profile configuration. For example, several studies have
compared offense variables such as rapists of adults,
rapists of children, and nonrapist sex offenders (Armentrout
& Hauer, 1978), murderers, arsonists, property offenders,
rapists, and child molesters (Quinsey et al., 1980),
exposers, assaulters, and rapists (Rader, 1977), incestuous
offenders and child molesters (Panton, 1979), and rapists of
adults, rapists of children, and non-violent child molesters
(Panton, 1978). Victim characteristics used in MMPI studies
have included age of victim (Groff & Hubble, 1984) and sex
of victim (Langevin et al., 1978), and offender
characteristics have included number of prior arrests
(McCreary, 1975a; McCreary, 1975b), sexual orientation
(Langevin et al., 1978), and relationship of offender to
victim (Groff & Hubble, 1984; Scott & Stone, 1986).
Table 1. Different Groups Used in MMPI Research with Sex Offenders
Study Categorization of Groups
Armentrout & Hauer (1978) Rapists of children, rapists of adults, non-rapist sex offenders
Carroll & Fuller (1971) Non-violent, violent, sexual
Erickson, et al. (1987) Rapists, child molesters, incest offenders, first time offenders, sex offender recidivists, non-sex offender recidivists
Groff & Hubble (1984)
Langevin, et al. (1978)
5
McCreary (1975a)
Incestuous fathers and incestuous stepfathers who abuse younger versus older daughters/stepdaughters
Homosexuals, transsexuals, bisexuals, homosexual pedophiles, heterosexual pedophiles, incestuous cases, exhibitionists, multiple deviants
Men convicted of indecent exposure'with no prior arrests, 1-5 prior arrests, more than 5 prior arrests
McCreary (1975b) Child molesters with no prior arrests, child molesters with one or more prior arrests
Rapists of adults, rapists of children, non-violent sexual molesters of children
Incest with daughters, molesters of female children
Quinsey et al. (1980) Murderers of family, murderers of non-family, arsonists, rapists, child
Rader (1977)
molesters, property crimes
Exposers, assaulters, rapists
It appears therefore that most of the more recent MMPI
research has compared group selected exclusively on the
basis of either offense type, victim characteristics, or
offender characteristics rather than systematically
including or controlling for more than one dimension. Hall
et al. (1986) and Hall (1989) criticize previous MMPI-sex
offender research for examining single variables, such as
offense variables, without controlling for or considering
variables, such as victim and offender variables, that
confound and interact.
One of the earliest studies to compare several
dimensions of sex offenders in relation to personality
factors was Langevin et al. (1978). In this study each of
their eight groups were categorized as preferring mature vs.
immature partners (age preference), males or females (sex
preference), and intercourse, rape, exposing, etc. (response
preference). However, this study's external validity is
compromised in that two of their groups (homosexual and
bisexual males preferring adult partners) would no longer be
considered relevant for studies of sex offenders, as they
are not presently considered anomalous characteristics
(American Psychiatric ~ssociation, 1987).
Hall et al. (1986) and Hall (1989) have perhaps
employed the most sophisticated method of group assignment
to date. These studies examined the utility of the MMPI in
differentiating between offenders who has assaulted male vs.
female victims, committed incestual vs. nonincestual
offenses, used physical force vs. nonphysical coercion,
raped vs.molested their victims, and assaulted older vs.
younger victims. Thus offender (relationship of offender to
victim), offense (type of coercion, rape vs. molestation),
and victim (sex, age) characteristics were simultaneously
compared for men who had sexually assaulted children. In
many previous studies, men who had sexually assaulted
children were defined as a homogeneous group and compared to
other groups exclusively defined by offense type, such as
rapists of adults, non-sex affenders, etc. (e.g. McCreary
1975a; Armentrout & Hauer, 1978; Quinsey et al., 1980).
Thus, this classification scheme (Hall et al., 1986; Hall,
1989) represents an improvement over the ones chosen in
previous MMPI research with sex offenders. However, only
one offender characteristic (relationship to victim) was
investigated.
Erickson, Luxenberg, Walbek, and Seely (1987) have also
emphasized the heterogeneity of personality characteristics
of sex offenders, and criticized previous research for
stereotypic descriptions of sex offenders, and the liberal
application of the article "the1' to describe the rapist, the
child molester, etc. This study (Erickson et al., 1987)
included more specific offender characteristics than any
previous MMPI studies. MMPI 2-point codes were collated
according to whether cases were intrafamilial or
extrafamilial, whether victims were adults or children, male
or female, whether subjects were recidivists or first time
offenders, and whether subjects were chemically dependent.
Recidivists were further divided as to whether their
previous offenses were sexual or nonsexual. Thus, both
offender and victim characteristics were investigated.
Unfortunately, offense characteristics, such as rape versus
molestation, were not included.
3. Definition of MMPI protocol validity
Studies relating the MMPI to sex offenders also differ
according to their definition of MMPI protocol validity. As
a result, different studies exclude different types of
offenders for different reasons, thus jeopardizing their
comparability. Several studies neglected to address 3
protocol validity (Anderson et al., 1979; Armentrout &
Hauer, 1978; Carroll & Fuller, 1971; Langevin, Paitich,
Freeman, Mann, & Handy, 1978). McCreary (1975a) omitted six
subjects on the basis of invalid profiles but did not define
the criteria for doing so. Some studies omitted subjects
solely on the basis of raw F scores above a particular
number (Quinsey et al., 1980; Rader, 1977). Still others
apply more than one criteria, for example L < 70, F < 100, K
< 70 (Erickson et al., 1987); L < 70, F < 85, K < 70
(Panton, 1979). An additional validity criterion method
used in some studies was the F-K index employing various raw
score cutoffs (Groff & Hubble, 1984; Hall et al, 1986).
Perhaps the most stringent criterion used for
eliminating invalid profiles in MMPI studies was that
employed by Hall et al. (1986). The criterion used in this
study excluded profiles with a raw score of F-K < 11 and a T
score of L < 70. Using this criterion 434 subjects, or two
thirds of their original sample, were deleted from analysis!
However, in a replication and extension of Hall et al.
(1986), Hall (1989) did not exclude subjects based on
~~invalidw profiles, and pointed out that Graham (1987, cited
in Hall, 1989) suggested that valid MMPI profiles may
mistakenly be discarded when employing overly conservative
approaches to MMPI validity.
Anderson et al. (1979) analyzed a group of 92 sex
offenders by a Q-type factor analysis and found that a
particular MMPI personality type was characterized by a high
F score, along with an elevated Sc score. Consequently,
they cautioned against discounting MMPI profiles due to
relatively high F scores.
Gynther (1962) found a significant relationship between
sexual crimes and high F scores. This, in addition to
Anderson et al.sf (1979) results, indicates that it may be
important to employ a more liberal approach in determining
MMPI profile validity in a sex offender population. Gearing
(1979) emphasizes that an F scale elevation above 80 (T
score) may actually be valid, especially with a prison
population. Gearing (1979) suggests that a good deal of
MMPI research with prisoners in general is compromised by an
I1overly rigid application of conventional validity criterial1
(p. 941). He therefore recommends that random profiles only
be discarded until further research produces dependable
indicators of faked prisoner profiles.
4. Statistical analysis of MMPI profiles
Most previous research with sex offenders has employed
univariate statistical techniques for data analyses (Carol1
& Fuller, 1971; Erickson et al., 1987; Groff & Hubble, 1984;
McCreary, 1975a; McCreary, 1975b; Panton, 1978; Panton,
1979; Rader, 1977; Scott & Stone, 1986). Several studies
used multiple t-tests without correcting for Type I error
(McCreary, 1975a; McCreary, 1975b; Panton, 1978; Panton, ,
1979; Rader, 1977), thus increasing the probability of
chance differences across groups being treated as genuine.
Occasionally the use of multiple t tests and F tests were
combined with a reduction of alpha to .O1 (Scott t Stone,
1986; Hall et al., 1986). Armentrout & Hauer (1978) did not
report the statistical methods used to analyze their data.
Even where more appropriate multivariate procedures are
used (e.g., MANOVA or discriminant function analysis) two
problems remain. First, the ratio of variables to number of
subjects is rarely sufficient to provide adequate
statistical power. ~iven the rule of thumb ratio of ten
subjects per variable, a minimum of 130 subjects would be
required to analyze the 13 MNPI scales alone, and an
additional 10 subjects for every additional independent
variable. Of the MMPI studies reviewed, only four meet this
requirement (Carol1 & Fuller, 1971; Langevin et al., 1978;
Hall et al., 1986; Erickson et al., 1987). Second, the
scales of the MMPI are moderately intercorrelated, and
highly multiply correlated, leaving little unique variance
for any single scale to discriminate offender types.
One way in which these problems can be circumvented is
to component analyze the MMPI scales into orthogonal latent
constructs, and then use the reduced number of component
scores for analyses. This would increase the power of
statistical analyses and alleviate the unique versus common
variance problem. Unfortunately, the cost in using this
procedure is limited interpretability of profiles at the
individual level, thereby reducing the clinical or practical
utility of results. However, given the numerous benefits of
component analysis for the purposes of the present study,
interpretability at the individual level was viewed as an
acceptable cost.
5 . MMPI scores reported
MMPI studies with sex offenders vary considerably in
which MMPI scores are reported. Frequently, standard MMPI
scale group means are reported in isolation (Carroll &
Fuller, 1971; McCreary, 1975a; McCreary, 1975b; Panton,
1978; Panton, 1979). Group means can be misleading because
they do not necessarily reflect modal profiles for a
particular sample (Butcher & Tellegen, 1978). In fact, Hall
et al. (1986) found that the Welsh code for their entire
sample did not correspond to any one individual profile in
that sample.
Rather than reporting group means, it is more useful to
report the frequencies or percentages of the different code
or profile types occurring in a sample. At a minimum, the
number or percentage of subjects in each group with a T
score above 70 should be reported for each of the standard
MMPI scales (for examples refer to Hall et al., 1986;
Langevin et al., 1978; Quinszy et al., 1980).
Ideally, the data should be presented to convey as much
information as possible. This would include reporting the
frequencies or percentages of order-independent two point
Welsh codes (e.g. scales 3-4 or 4-3 are elevated), of the
frequencies or percentages of subjects in each group with a
T > 70 for each of the clinical scales, in addition to the
group means. This would facilitate comparison of results
across studies.
MMPI results in sex offender studies
Combined, the methodological problems discussed above
make MMPI/sex offender research results difficult to compare
and interpret. Therefore, rather than reporting an
extensive list of MMPI/sex offender research findings, the
following discussion will focus on the general findings,
with particular reference to the methodological issues '
previously addressed.
Despite wide variations in methodological procedures,
studies relating MMPI profiles to sex offenders have tended
to converge on the K-corrected 4-8/8-4 profile as the mean
profile for sex offenders in general (Armentrout & Hauer,
1978; Hall et al., 1986; Panton, 1958; Panton, 1978; Quinsey
et al., 1980; Rader, 1977; Pwenson & Grimes, 1958). The 4-
8/8-4 profile was reported for various sex offense
categories including rapists of children (Armentrout &
Hauer, 1978; Panton, 1978), rapists of adults (Armentrout &
Hauer, 1978; Panton, 1978; Rader, 1977), child molesters
(Erickson et al., 1987; Hall et al., 1986; McCreary, 1975a),
exposers (McCreary, 1975b; Rader, 1977), and multiple
deviants (~angevin et al., 1978).
However, although the 4-8/8-4 profile appears to be the
most frequently occurring profile in various groups of sex
offenders, it is sometimes the case that relatively few
individuals produce this profile. For example, Hall et al.
(1986) reported that although the modal profile for their
sample was 4-8/8-4, only 7.2 % of their subjects exhibited
this profile. Armentrout & Hauer (1978) found that although
4-8/8-4 was the modal profile for their sample of rapists,
it occurred for only three of the 17 individuals. As
previously mentioned, very often MMPI studies report group
means on the standard MMPI scales without reporting the
frequency or percentages of individuals obtaining scale
elevations or 2-point codes.
Studies have also frequently reported finding no
significant differences of MMPI profiles between various
groups of sex offenders. Quinsey et al. (1980) compared
several groups with a wide range of offense characteristics,
including rapists and child molesters, and found that Itthe
extreme similarity of the offender groupst MMPI scores is
the most salient result of this studyw (p. 414). Groff and
Hubble (1984) found no systematic MMPI profile differences
between incestuous fathers and stepfathers. Armentrout and
Hauer (1978) reported that rapists of adults and non-rapist >
sex offenders bothexhibited the same modal MMPI profile (4-
8/8-4). Hall et al. (1986) and Hall (1989) concluded that
sex offenders are not characterized by a specific MMPI
profile configuration. Erickson et al. (1987) reported that
their findings "do not support descriptions of any MMPI
profile as typical of any sort of sex offenderu (p. 569).
Several studies have also found that sex offender MMPI
profiles do not differ significantly from those of offenders
convicted of other crimes (Hartmann, 1967; Perdue & Lester,
1972; Persons & Marks, 1971). Erickson et al. (1987) found
that MMPI profiles of sex offenders were remarkably similar
to those of arsonists, murderers, and property offenders.
Quinsey et al. (1980) found no significant differences
between individuals convicted of sexual offenses versus
those convicted of other offenses. Quinsey et al. (1980)
also found that the mean K-corrected MMPI clinical scales
for their total sample correlated highly with samples from
other studies of hospital narcotics addicts (r=.91), heroin-
using male veterans (r=.85), and personality disordered male
veterans (r=. 85) . Finally, several studies reported that various sex
offender groups had essentially normal profiles. Carroll &
Fuller (1971) found that the MMPI profiles of violent and
sex offender groups did not differ significantly from the
normal, general, non-incarcerated population. Rader (1977)
reported that exhibitionists had normal MMPI profiles. >
Langevin et al. (1978) noted that I1most subjects have T
scores below 70, that is, they score within normal limitsM
(p.232). Anderson et al. (1979) conclude after a review of
MMPI studies that MMPI profiles of sex offenders reflected
minimum psychopathology. Scott and Stone (1986) found that
their groups of incestuous fathers and stepfathers produced
mean profiles in the normal range.
Thus, it appears that MMPI research with sex offenders
has produced equivocal results to date. In general, sex
offenders most often exhibit profiles that do not
differentiate them from other offender populations, and
20
indeed frequently produce clinically normal profiles.
Undoubtedly part of the problem with interpreting the
results of studies stems from the methodological issues
discussed earlier. However, even studies that are more
methodologically refined (e.g. Hall et al., 1986; Hall,
1989; Erickson et al., 1987) produced negative results.
Indeed, it appears that when attempts are made to control
for variables that may confound or interact, the MMPI
becomes less useful in discriminating various types of sex
offenders.
A second potential reason for equivocal MMPI results is
the relative infrequency among sex offenders of the very
clinical syndromes (Axis I of the DSM-111) best captured by
the MMPI (Finkelhor, 1978; Maisch, 1972; Meiselman, 1978).
For example, Groff & Hubble (1984) found that nearly half of
their sample of incestuous offenders fell within Goldberg's
(1972) sociopathic category, indicating a variety of
personality disorders, and conclude that this finding
suggests that incestuous offenders "were more likely to
exhibit characteristics of personality disorders rather than
neurotic or psychotic symptoms.It (page 472). Thus, sex
offenders may be more likely to harbour personality
disorders (i.e., Axis I1 of the DSM-111), and attempts to
assess them should use an instrument that is more sensitive
to those disorders. One such instrument is the Millon
Clinical Multiaxial Inventory, a self-report measure
specifically designed to assess both Axis 1 and personality
disorders. Thus, the centrr-1 thesis of the present research
is that the MCMI will better capture the psychological
characteristics of sex offerders (defined simultaneously
along offense, offender, and victim dimensions) than will
the MMPI. The next section will describe the MCMI and end
with the specific goals of the present study.
The Millon Clinical Multiaxial Inventorv
The MCMI is a relatively recent (Millon, 1977) self-
report inventory, and is a direct operationalization of
Millon's theoretical system and taxonomy (Millon, 1969,
1981). The MCMI includes 8 basic personality styles
(schizoid, avoidant, dependent, histrionic, narcissistic,
antisocial, compulsive, passive-aggressive), 3 pathological !
personality syndromes (schizotypal, borderline, paranoid), 6
symptom disorders scales of moderate severity (anxiety,
somatoform, hypomanic, dysthymia, alcohol abuse, drug abuse)
3 symptom disorder scales of extreme severity (psychotic
thinking, psychotic depression, psychotic delusions), plus 2
additional correction scales which provide a means to \
identify and adjust possible test taking distortion
(Mitchell, 1985) . A major advantage of the MCMI over the MMPI is its
relationship to the current diagnostic nomenclature
(Widiger, 1985). The MMPI scales and profile codes are
based on the antiquated DSM-I. Current research and
clinical practice has shifted from an emphasis in the
diagnosis of psychiatric syndromes (DSM-111, Axis I), to the
diagnosis of personality disorders (DSM-111, Axis 11). The
MCMI is the self-report inventory most closely related to
Axis I1 diagnoses (Widiger, 1985).
More recently, however, Widiger, Williams, Spitzer, &
Frances (1985) question the use of the MCMIts ability to
measure DSM-I11 Axis I1 disorders. They point out that
there has been no published empirical research addressing
the relationship between the MCMI and DSM-111. Millon
(1985b), in response to ~idiger et a1.I~ (1985)
implications, argues that the I1MCMI was never intended to be
a measure of DSM-111" (p. 379), but was developed to be as
consonant as possible with DSM-I11 classifications.
Millon (1985b, 1986) further argues that his
borderline, schizotypal, and narcissistic personality
syndromes are conceptually equivalent to the corresponding
DSM-I11 categories. The MCMI and DSM-I11 correspond only to
the extent that the conceptual and clinical categories of
Millon's theory and the DSM-I11 overlap (Millon, 1986).
Indeed, Widiger & Sanderson (1987) in a study using the MCMI
with 53 psychiatric hospital inpatients, found better
convergent validity for the DSM-I11 disorders that were
consistent with Millonts typology (avoidant and dependent
personality disorders) than for disorders that were
inconsistent (antisocial and passive-aggressive).
There have been relatively few empirical studies using
the MCMI to date. The first study to investigate the
relationship of the MCMI to an offender sample was conducted
very recently by McCormack, Barnett, & Wallbrown (1989).
One thousand two hundred MCMI offender profiles were
randomly divided into two groups of 600, and subjected to a
principal components factor analysis. Successful cross
validation was achieved, as the four factors derived in both
groups were similar.
Purpose of present study
The purposes of the present study are:
1. To categorize sex offenders according to more offender,
offense, and victim characteristics than previous MMPI
studies.
2. to compare the predictive validity of the MMPI and MCMI
with respect to offender, offense, and victim
characteristics.
3. to evaluate the common and unique aspects of offender
characteristics, the MMPI, the MCMI for predicting offense
and victim characteristics.
CHAPTER 11
METHOD
Subj ects
Subjects were 127 male sex offenders who had been
referred to Forensic Psychiatric Services, Victoria, for
predisposition assessment or treatment between July 1, 1986
and December 31, 1989. All those subjects who had completed
nonrandom MMPIts (~earing, 1979) and MCMI1s, and had a grade
six or higher level of education (Hathaway & McKinley, 1967)
were included in the sample. The Carelessness scale
(Greene, 1978) was used to determine random MMPI profiles,
employing a cutoff score of five of the 12 item pairs
(Greene, 1980), which would indicate that the subject was
contradicting himself to the extent that responses were
random, or misunderstood. An MCMI Validity Index score of
zero indicated a nonrandom profile (Millon, 1983).
Procedure
Offense history data were obtained from the clinical
files of sex offenders at Forensic Psychiatric Services
outpatient clinic. Those subjects who completed nonrandom
MMPI Group form and MCMI between July, 1986 December, 1989
were included for analyses. In addition, a description of
the offender, offense, and victim characteristics used for
analyses follows. victim and offense characteristics were
combined for statistical analyses and are jointly termed
offense variables. It should be stressed that this sample
of sex offenders is somewhat posively skewed with respect to
age of victim, i.e., there was a relatively high proportion
~f offenses against individrals under age 16. Also, because
the subjects were attending an outpatient clinic, by
definition the offenses were not serious enough to warrant
remand in custody, and the subjects were not incarcerated at
the time of assessment.
Offender Variables:
Age in years at initial offense - 11-20, 21-30, 31-40, 41-70 Employment - unemployed, retired, skilled labour (professional, sales, entrepreneur), unskilled labour
Education - high school graduate or higher, grade eleven or less
Marital status at the time of offense - single, divorced or separated, married or cohabitating, remarried
History of physical!abuse - history versus no history of physical abuse
History of sexual abuse - history versus no history of sexual abuse
History of criminal offenses - presence versus absence of criminal record
Substance abuse - drugs (narcotics, prescription, or one of these in combination with alcohol), alcohol, no substance abuse
Offense and Victim Characteristics:
Number of victims - one victim versus more than one victim Relationship of offender to victim - intrafamilial (biological father, step-father, uncle, grandfather, sibling) versus extrafamilial (friend, stranger)
Type of offense - relatively more serious (digital, vaginal,
or anal penetration, oral sey: to or by victim), relatively less serious (hands off, e.g:, exposing, obscene phone calls; fondling or masturbation to or by victim)
Highest number of repeats with any given victim - once only per victim, 2-19 incidents, 20 or more incidents
Youngest age of any given victim at the time of the first offense -ten years or younger, eleven years or older
Other variables were initially coded but excluded due
to restricted range, defined as 70 percent or more subjects
belonging to any one category. These included sexual
orientation (78% heterosexual, 8% homosexual, 5% bisexual),
sex of victim (74% female, 13% male, 13% both), and physical
coercion or aggression involved in offense (71% no, 20% yes,
9% unknown). In addition, marital status of offenders'
parents was coded but was excluded when preliminary analyses
indicated that it was unrelated either singly or jointly to
any other variables:
Outline of Statistical Analyses
Two classes of analyses were used to evaluate the data.
The first used procedures typical of previous research
investigating the MMPI profiles of sex offenders, but
followed the advise of ~earing (1979) to investigate several
different quantifications of MMPI protocols. Of course,
both the MMPI and MCMI were used and compared. Examples of
the first class of analyses include comparison of high-point
profiles, order independent conjoint 2-point codes, and
group mean profiles.
The second class of analyses focused on the overall
relations among MMPI, MCMI, offender, and offense variables
and proceeded as follows:
1. Data reduction. The offender and offense variables were
reduced using cluster analysis, and the MMPI and MCMI scales
were component analyzed.
2. Analvsis of airw wise relations. Each pair of variable
sets was analyzed using MANOVA, discriminant function, or
canonical analysis where appropriate. Examples of pairwise
relations include offender/offense relations,
personality/offender relations, and personality/offense
relations. These analyses might seem redundant given step 3
of the analysis described below in which multiple-set
relations are investigated. They were included for two
reasons. First, previous research on sex offenders has
investigated pairwise set relations (e.g., MMPI/offender
relations) and could be compared with the current results
only if similar procedures were used here. The second
reason was that the results of multiple-set analysis are
easier to interpret if the simpler pairwise relations are
understood, much like partial correlations are more
interpretable if lower-order correlations are understood
first.
3. Multiple-set relations. In this analysis offender
clusters and personality components were considered as sets
of predictor variables and offense clusters as a set of
criterion variables. Multiple set correlation (see Cohen,
1982) was used to evaluate the common and unique aspects of 1
each predictor set in "e~plaining~~ the criterion set. More
is said about this technique below.
Analvsis of 2-point codes and mean profiles
The following is a discussion of MMPI and MCMI results
analyzed in the same manner as in most previous sex offender
research. The mean MMPI 2-point code for the entire sample
was 2-4/4-2, which was also the modal 2-point code. The
MCMI mean 2-point code for the total sample was 3-A/A-3,
although the modal profile for the sample was D-A/A-D. On
both instruments the mean 2-point codes were not clinically
elevated. Table 2 presents the MMPI and MCMI means for each
scale for the total sample.
Although 2-4/4-2 was the mean and modal MMPI profile
for the total sample, only eight subjects, or 6.3% of the
sample, exhibited this profile. Table 3 illustrates the
frequencies and percentages of subjects obtaining the most
common order-independent 2-point codes occurring in the
present sample. Clearly the sample produced heterogeneous
MMPI profiles, indicating that no particular 2-point code
captures the sex offenders in this sample. Scales 2 and 4
were also the most frequently elevated scales in the sample.
Table 4 presents the frequency and percentage of subjects
obtaining a T-score 70 or higher on each of the MMPI scales.
The MCMI order-independent 2-point codes and single
scale elevations are presented in Table 5. The D-A/A-D 3
(Anxiety and Dysthymic) scales comprised the 2-point codes
for 21 subjects, or 16.5 percent of the total sample.
Twenty percent, or 25 subjects, had scale 3
(dependent/submissive) elevated above baserate (BR) of 75
and as their highest scale elevation. There were 10
subjects (8%) whose profiles yielded no scale elevations
above a BR of 75, versus 19 subjects (15%) with MMPI
profiles with all scales below a t-score of 70. Table 6
presents the frequency and percentage of subjects obtaining
a BR score of 75 or higher on each of the MCMI scales.
The three most frequently occurring order-independent
Table 2. MMPI and MCMI Means for Total Sample
MMPI Scale Mean MCMI Scale Mean
Table 3. Order Independent IqYlPI 2-Point Codes
T a b l e 4 . Frequency and Percentage of subjects in Total Sample with MMPI T-scores >69
Scale N %
T a b l e 5. Order Independent MCMI 2-Point Codes and Highest single Scale lev at ions (above BR 75)
Scale N %
codes on the MCMI account for 31.5 percent of the total
sample, whereas the three mast frequently occurring order-
independent MMPI codes only account for 16.5 percent of the
sample. Thus, it appears that the sample is more
homogeneous on the basis of MCMI than on MMPI profiles.
Table 6 shows the frequency and respective percents of
subjects with a BR above 74 on each MCMI scale.
As discussed in the ~ntroduction, previous MMPI studies
with sex offenders frequently employed multiple t-tests to
compare the mean MMPI scores for each group of interest,
resulting in an increase of Type I error. In the present
study there are 13 offender/offense variables and 33
MMPI/MCMI personality variables. Therefore, rather than
resorting to a confusing and undoubtedly erroneous welter of
univariate analyses the data were first reduced and then )
analyzed using to multivariate procedures.
Data reduction -- Offender characteristics Table 7 presents a table of contingency coefficients,
which highlights significant correlations among the offender
characteristics. The following is a list of offender
variable abbreviations and their explanations:
AGE - Offender age at initial sexual offense EDUC - Education level EMPLOY - Employment MARITAL - Marital status at time of offense PHYSAB - History of previous physical abuse SEXAB - History of previous sexual abuse SUBAB - Substance abuse CRIM - Previous criminal record
Table 6. Frequency and Percentage of Subjects in Total Sample with MCMI BR Scores > 75
S c a l e N %
Table 7. Contingency coefficients Among Offender Variables
AGE EDUC EMPLOY MARITAL PHYSAB SEXAB SUBAB CRIM AGE 1.00
EDUC .05* 1.00 EMPLOY .39*** .19 1.00
MARITAL . 4 5 . 2 2 .30 1.00 PHYSAB . 2 6 .11 . 1 8 .25 LOO**
SEXAB .19 .17 .12 .31* 1.00 SUBAB .28 .37 .20
CRIM .17 .09 .24 .12 *26** .32 .13 1.00
Clearly one would not expect that sex offenders
represent a homogeneous group -- a point made throughout the sex offender literature. Therefore, the possibility that
the current sample of offenders could be grouped into types
that display systematic differences in offender
characteristics was investigated using cluster analysis.
Unfortunately, there is no universal agreement on the "bestu
clustering algorithm, so several were attempted including
hierarchical and partitioning methods as well as a recently
designed nonparametric procedure (Brinkhurst, 1988). These
analyses converged on a three cluster solution in which
groups differed in age at time of initial offense,
employment, and marital status at time of offense. The
remaining offender variables of education, history of
physical and sexual abuse, and criminal record, did not I
differentiate the initial three groups and were submitted to
a second cluster analysis. The results produced a two
cluster solution. A description of the results of these
analyses is presented below.
Three cluster solution
The first cluster (N = 44) is the youngest group of the
three, and will therefore be referred to as the "young
groupat. Individuals who comprise this cluster are likely to
be 30 or younger, employed at unskilled labour, 'and single.
The second cluster (N = 48) are primarily unemployed (the
"unemployed groupw) and are most likely to be between the
ages of 31-40. They are equally likely to be divorced,
married, or single. The third cluster (N =35), and oldest
of the three groups were most likely employed as skilled
laborers, and married ("married groupI1 ) . Two cluster solution
Offenders that comprise the first cluster (N = 57) have
a relatively unremarkable history in that they are likely to
have no history of physical or sexual abuse, and no previous
criminal rec,ord. On the other hand, offenders in the second
cluster (N = 37) do have a remarkable history in that they
were likely to have been physically and sexually abused, and
have a previous criminal record. They were also less
educated.
Data reduction -- Offense characteristics Table 8 presents a table of contingency coefficients
!
among the offense characteristics. The following is a list
of offense abbreviations and their explanations:
NOVIC - Number of victims FAMIL - F elation ship of offender to victim OFFNS - Type of offense REPEAT - Highest number of repeats with any victim AGEFIR - Youngest victimls age at time of first offense
Cluster analysis of these offense characteristics
produced a three cluster solution on which all of the
offense variables contributed to group differences. The
first cluster (N = 54), referred to as the "intrafamilial
group81, were more likely to offend against one intrafamilial
victim, the victim was typically 10 years old or younger,
Table 8. Contingency coefficients Among Offense Variables
NOVIC FAMI L OFFNS REPEAT AGEFIR
NOVIC FAMIL OFFNS REPEAT AGEFIR 1. ooo** .403 1. OOO** .072 .469** 1. OOO** .I93 .509** .404** 1. OOO* .089 .354 .325 .295 1.000
Table 9. Contingency Coefficients Among Offense and Offender Variables
AGE EDUC EMPLOY MARITAL PHYSAB SEXAB SUBAB CRIM
NOVIC OFFNS REPEAT AGEFIR
the number of repeat offenses was 20 or more, and the
offense was of a relatively more serious nature. The second
cluster (N = 52), or tfextrafamilial grouptt, were also likely
to have one victim only, but the victim was likely to be
extrafamilial, age 11 or older, and the offense was
typically one time only and of a relatively less serious
nature. The third cluster (N = 21), or "chronic groupu were
likely to offend against more than one victim, and the
victims tended to be both intrafamilial and extrafamilial.
The victims were ten or younger and there were typically
between 2-19 offense repeats per victim. The offenses
themselves were of a relatively more serious nature.
Data reduction - Personality variables MMPI T-scores were component analyzed followed by
normalized varimax rotation. The resulting four-factor 9
solution and corresponding variances accounted for are shown
in Table 10 where it can be seen that most of the scales are
factorially complex, i.e., they load on more than one
factor. In naming the factors (see below), labels were
chosen that reflected those scales that appeared to be
uniquely determined by the underlying construct.
Factor 1 -- Impulsive factor: This factor has its
maximal positive loading on Scale 9 (Ma), and reflects
energetic, impulsive, irritable, grandiose, and moody
characteristics.
Factor 2 -- Somatic factor: Maximal positive loadings
Table 10. MMPI Rotated Factor Matrix (N=127)
Rotated Loadings 1 -. 440
.622 -.491 .017 .009 .022 .493
-.071 .482 .342 .600
*. 899 -.lo2
% of total 19.991 22.179 19.776 14.359 variance explained
* Loading greater than .80
are found.on Scales 1 (Hs) and 3 (Hy), indicating a tendency
to develop somatic problems, and a demanding, complaining
personality style. Other characteristics of these scales
are attention-seeking, hostility, repression, denial,
manipulativeness, and lack of insight.
Factor 3 -- Introversi~n factor: This factor loaded
primarily on Scale 0 (Si), suggesting social introversion,
shyness, overcontrol, lethargy, and withdrawal.
Factor 4 -- Sensitivity factor: Scale 5 (Mf) is the
salient feature of this factor, indicating passivity,
sensitivity, and conflicts over heterosexual behaviour.
Interestingly, Scales 2 (D) and 4 (Pd), which comprise
the mean and modal 2-point code for the total sample, did
not load exclusively on any one factor. In fact, Scale 4
loaded almost equally on three factors, indicating that 1
Scale 4 characterizes several features of the criminal
population in general, such as impulsivity, apathy, and
social maladjustment. Also, there may be a more complex
relationship of this scale to other MMPI scales than has
been hypothesized in previous research. This may partially
explain the inconsistent results in previous MMPI studies
with sex offenders.
Table 11 presents the corresponding component analysis
of the MCMI scales.
Factor 1 -- Bizarre/withdrawn: this factor is bipolar with the positive pole defined by loadings on Psychotic
Table 11. MCMI Rotated Factor Matrix (N=127)
Rotated Loadings
Schizoid Avoidant Dependent Histrionic Narcissistic Antisocial Compulsive Passive-Aggressive Schizotypal Borderline Paranoid Anxiety Somatoform Hypomanic Dysthymic Alcohol Abuse Drug Abuse Psychotic Thinking Psychotic Depression Psychotic Delusions
Percent of Total 27.502 ' !
Variance Explained
Table 12. Pearson Correlation Matrix of MMPI and MCMI Factors
MMPI Factors 1 2 3 4
MCMI Factors 1 .421 .067 .709 .130 2 .661 -.204 -.342 -.010 3 .152 .270 .I18 .270 4 .078 -.178 .019 -.236
Depression (. 881) , Avoidant (. 848) , and Schizoid (. 818) . The negative pole includes loadings on compulsive (-.676)
and Narcissistic (-.526). Characteristics indicated by the
positive pole of this factor are behavioral apathy,
depression, interpersonal indifference, social isolation,
and alienated, devalued self image. The negative pole of
this factor involves an absence of motivation to conform to
the expectations of others, and of gregariousness/
sociability. Because this factor is bipolar, one would
expect those with low factor scores to evidence the opposite
of this factor's salient features. Therefore, individuals
scoring low on this factor would be expected to score high
on the Narcissistic and compulsive scales.
Factor 2 -- Sensation-seeking: This factor was defined ,
primarily by loadings on Drug Abuse (.868), Hypomanic
(.832), and ~istrionic (.804). The common themes of this
factor indicate impulsivity, emotional lability,
irritability, history of drug abuse, sociable/gregarious
self-image, immature stimulus seeking behaviour, and short-
sighted hedonism.
Factor 3 -- Somatoform: This factor is characterized
by positive loadings on Somatoform (.922), Anxiety (.882),
Borderline (.840), and Dysthymic (.832). Factor 3 is
bipolar, with its negative pole defined by a loading on
~ntisocial (-495). Individuals scoring high on this factor
are likely to evidence anxiety, somatic tension, dysphoria,
massive interpersonal ambivalence, preoccupation with
feelings of discouragement or guilt, as well as a lack of
antisocial behaviour. Recall from the previous discussion
that 16.5% of the total sample evidenced ~nxiety and
Dysthymic 2-point codes, both of which load on this factor.
Factor 4 -- Paranoid: This factor was defined by
loadings on Psychotic Delusions (.869) and Paranoid (.776),
indicating paranoid behaviour and thinking with the
associated features of suspiciousness, and hostile acting
out.
Table 12 presents the Pearson correlations among the
MMPI and MCMI factors. Clearly there is overlap between
the two instruments even at the level of latent variables.
For example, MMPI factor 3 (Introversion) is correlated with ,
MCMI factor 1 (Bizarre/withdrawn), and MMPI factor 1
(Impulsive) is correlated with MCMI factor 2 (sensation-
seeking). MMPI factor 3 indicated social introversion,
shyness, lethargy, while MCMI factor 1 indicated behavioral
apathy, depression, interpersonal indifference, and social
isolation. MMPI factor 1 reflected energetic, impulsive,
irritable, grandiose, and moody characteristics. MCMI
factor 2 indicated impulsivity, emotional lability,
irritability, and sociable/gregarious self-image. The
characteristics described by the correlated factors on the
MMPI and MCMI appear to be behaviorally and cognitively
similar.
Pairwise set relations -- Offender clusters and MMPI/MCMI factors
As previously stated, the offender and personality
(MMPI and MCMI factors) variables in this study are analyzed
and interpreted as independent variables or predictors. The
dependent variables, or criteria, are the offense variables.
Before exploring the relationships between the predictor and
criterion variables, relationships that exist between the
two groups of predictor variables will be investigated.
A factorial MANOVA was used to determine the
relationship of personality factors to the two and three
cluster solutions for the offender variables. The analysis
therefore consisted of testing the main effects and
interaction of the offender cluster solutions on
personality. This analysis yielded a significant main
effect for the two cluster solution of the offender
variables (Wilks' Lambda = .807, F(8,114) = 3.398, p <
.002). The offender group with an abusive history was
significantly more likely, on average, to have a higher
score on the MMPI Impulsive and Sensitivity factors
(F(1,121) = 14.337, p < .001; F(1,121) = 5.117, p < .03
respectively), and the MCMI Bizarre/withdrawn factor,
-F(1,121) = 10.991, p < .003). The three cluster solution
for the offender group, which produced the young,
unemployed, and married groups was not significantly related
to the personality factors (Wilksf Lambda = .862, F (16,228)
= 1.100, p < .36). Additionally, the interaction between
the offender and personality variables was not significant
(Wilksf Lambda = .835, F(16,228) = 1.345, p < .17).
Pairwise set relations -- Offender and offense clusters This section and the one to follow will focus on the
relationship of each of the two predictor groups separately
with the offense, or criterion groups. The offender
clusters (predictors) and offense clusters (criterion) are
categorical variables leaving open several possibilities for
analysis (e.g., log-linear analysis, logit/probit analysis).
The approach chosen here was to code the offender and
offense clusters to produce meaningful single degree
contrasts among groups and then submit those contrasts to a
canonical correlation analysis. For example, the three
offense clusters were coded into two contrasts representing
(1) a comparison between the intrafamilial group and the
chronic group, and (2) a comparison between the extrafailial
group and the average of the intrafamilial and chronic
groups. Similarly, the offender cluster solutions were
coded to reflect the main effects of the two and three
cluster solutions and their interaction.
Neither the main effect for the two cluster solution of
offender variables (nonabusive versus abusive history) nor
the interaction were significant (p < .32 and p < .lo,
respectively). The main effect of the three cluster
solution was significant (Wilkst Lambda = .769, F(4,240) =
8.428, p < .001) indicating that intrafamilial offenses
were, on average, more likely to be committed by individuals
belonging to the married cluster, whereas the chronic
offenses were more likely to be committed by individuals
belonging to the unemployed or young clusters (F(2) =
13.371, p < .001). Additionally, the extrafamilial offenses
were more likely to be committed by individuals belonging to
the young offender group, while the intrafamilial and
chronic offenses were more likely to be committed by the
married offender group (F(2) = 8.022, p < .001). Overall,
it appears that the young offender group is much more likely
to commit extrafamilail offenses, and the married group is
far more likely to commit intrafamilial offenses but rarely )
commit chronic or extrafamilial offenses.
Pairwise set relations -- Personality factors and offense clusters
The previous section described the relationship between
the offender predictors and the offense criterion. This
section will focus on the relationship between the
personality predictors and the offense criterion. First,
separate discriminant function analyses were carried out
using the 13 MMPI and 20 MCMI scale scores. These analyses
emulated the pairwise relations investigated in some
previous sex offender research. Next, the four MMPI and
four MCMI factors were separately analyzed to see if they
could discriminate the contrast coded offense groups
described in the previous section. Finally, because the
MMPI and MCMI personality Eactors are intercorrelated (see
Table 12), a discriminant function analysis was done on each
of the two sets of personality factors with the other
partialled out. In other words, the MMPI was analyzed with
the MCMI partialled out, and vice versa for the MCMI, to
explore the unique relationship of each of these instruments
with the criterion, or offense groups.
The first part of these analyses, a discriminant
function analysis separately on all MMPI and MCMI scale
scores, produced no significant results (Wilks' Lambda =
.760, F(26,224) = 1.27, p < .20; Wilksl Lambda = .696,
F(40,210) = 1.043, p < .40, respectively). That this
analysis does not even approach significance can be
attributed in part to the fact that the ratio of subjects to
variables is too high, thereby diminishing the power of the
test. As discussed previously, this circumstance has been
the downfall of many previous MMPI studies with sex
offenders.
When the same analyses are done on the reduced number
of personality factors (the number of variables drops from
33 to 8) and with offense groups coded into meaningful
contrasts a different scenario emerges. The MMPI factors
remain not significantly related to the offense contrast
codes (Wilkst Lambda = .90, F(8,242) = 1.638, p < .12).
However, the M C M I factors alone are significantly related
(wilksl Lambda = .874, F(8,242) = 2.115, p < .03), and both
discriminant functions are significant. The first
discriminant function, contrasting the intrafamilial and
chronic offense clusters was significantly discriminated by
the bizarre/withdrawn M C M I factor (Bartlett s x2 (8) = 16.56.
p < .03), with individuals committing the chronic offenses
obtaining a higher mean factor score on the
bizarre/withdrawn M C M I factor. The second discriminant
function indicated that individuals belonging to the
extrafamilial cluster were significantly more likely to
obtain a higher score on the M C M I paranoid factor than the
average of the intrafamilial and chronic groups (Bartlett
x2(3) = 7.741, p < .05).
The final analysis relating the personality factors and
the contrast coded offense clusters utilized a partialling
procedure. When the MMPI is analyzed with the M C M I
partialled out, a significant relationship emerges between
this instrument and the contrast coded offense clusters
(wilkst Lambda = .869, F(8,234) = 2.127, p < .05). One
discriminant function was significant, with the somatic MMPI
factor discriminating individuals belonging to the
intrafamilial offense cluster from those belonging to the
chronic offense cluster (Bartlett x2 (8) = 17.146, p < .O5) . The chronic offense group obtained, on average, a higher
score on the somatic MMPI factor than did the intrafamilial
group.
A discriminant function analysis of the MCMI factors
with the MMPI factors partialled essentially replicated'the
results of the unpartialled analysis (wilksl Lambda = .844,
F(8,234) = 2.589, p < .01). Both discriminant functions
were significant (Bartlettfs x2 (8) = 20.76, p < .01; x2(3)
= 8.386, p < .03). Individuals belonging to the chronic
offense clusters were significantly more likely to obtain a
higher score on the MCMI Bizarre/withdrawn factor and a
lower score on MCMI Somatoform factor than those belonging
to the intrafamilial offense cluster. ~ndividuals belonging
to the extrafamilial offense cluster were again more likely
to obtain a higher score on the MCMI paranoid factor than
were the intrafamilial and chronic offense groups. >
~ultiple-set relation -- Offender, ~ersonalitv and offense variables
The focus of the analysis was to predict offense
characteristics (the criterion) from offender and
personality variables (the predictors). The procedure used
was multivariate set correlation (Cohen, 1982).
Conceptually, multiple-set correlation is a generalization
of multiple regression/correlation in which sets of
predictor variables are used to predict a set of criterion
variables rather than the usual relating of single
indicators of each. Set correlation provides indices of
association analogous to those provided in MCR analysis, for
example the multivariate multiple correlation coefficient,
the multiple partial correlation and the multiple
semipartial correlation. Similarly there are multivariate
regression weights analogous to the partial regression
weights associated with standard multiple regression
procedures. This procedure provides numerous analytic
possibilities (see Cohen, 1982) and is ideal for the current
situation because it provides information concerning the
association between the common aspects of
predictor/criterion sets as well as between their unique
aspects. Details on computing F ratios and degrees of
freedom are given in Cohen, 1982, p. 318-322.
Many significance tests arise in multiple-set
correlation so protection of Type I error inflation is an I
important issue. Cohen (1982) recommends a hierarchical
procedure in which relations among larger sets of variables
should be significant before testing separate or partialled
subsets. For example, in the current context the overall
association between offender characteristics, personality
factors, and offense clusters should be statistically
significant before testing offender/offense relations with
personality partialled or personality/offense relations with
offender characteristics partialled.
The results are shown in Tables 13, 14 and 15. These
tables are identical with respect to the progressive
partialling of predictor sets. That is, moving down the
first column of each table indicates that the sets of
predictors were considered si~.ultaneously before proceeding
to investigate subsets of predictors from which others had
been partialled. The tables differ only in their criterion
variables. In Table 13 both contrasts among the offense
groups ( i t intrafamilial vs. chronic groups, and
extrafamilial versus the average of the intrafamilial and
chronic groups) were analyzed as a criterion set, whereas in
Tables 14 and 15 each contrast is analyzed with the other
partialled out. Table 16 summarizes the entire analysis.
Some possible sets are missing from the tables. For
example, the set representing the interaction between
offender clusters proved unreliable in all analyses and
therefore is omitted from the tables. Interactions among ,
offender groups and personality factors could have been
investigated but would have required constructing an
additional 40 variables; a situation that would have been
largely uninterpretable, and essentially meaningless given
the resulting ratio of subjects to variables.
All higher order relations are statistically
significant, meeting the criteria for Type I error
protection and allowing the discussion to focus on the
relationships among partialled single-set predictors to
partialled criteria (i.e. the last four rows of Tables 14-
15, and the last four rows and two columns of Table 16).
Table 13. Set Correlation ~nalysis of Offender and Personality Variables Related to Coded set of Offense Variables
Independent Variable R~ F df P
o1 - two cluster solution of offender variables O2 - three cluster solution of offender variab1.e~ P1 - MMPI four factors P2 - MCMI four factors Y1 - Intrafamilial group versus chronic group Y2 - Extrafamilial vs. average of Intrafamilial and Chronic
groups R2 - Squared multivarihte (partial, semipartial) correlation
Table 14. Set correlatirn ~nalysis of Intrafamilial Group Versus chronic Group
Independent Variable R~ F df D
Table 15. Set Correlation Analysis of the Extrafamilial Group Versus the Average of t h e Intrafamilial and Chronic Groups
Independent Variable R~ F df P
Table 16. Summary of Set Correlation Analysis
Independent variable Dependent Variable
Y1u2 Yl'~ 2 y2.y1,
Essentially, those results corroborated the results
discussed in the previous three sections with a few
interesting departures. The two cluster solution for the
offender variables with everything else partialled out
differentiated the extrafamilial offense group from the
average of the intrafamilial and chronic offense groups
(multivariate R2 = .O57, p < .OX) . The extrafamilial
offense group was less likely to have had an abusive history
than either of the other offense groups, who did not differ
from one another (multivariate R2 = .044, ns. )
The relationship between the three cluster solution of
offender variables and the contrast coded offense variables
corroborated the unpartialled results reported above that
found married offenders more likely to commit intrafamilial
offenses (multivariate R2 = .l98, p < .001) . However,
unlike the findings above, there is no significant
difference between the extrafamilial offense group
contrasted with the average of the intrafamilial and chronic
offense groups in terms of which offender group they belong
to.
The relationship between the MMPI factors and the
offense variables was not significant when analyzed without
partialling. However, after partialling the MMPI Samatic
factor differentiated the intrafamilial and chronic offense
groups (multivariate R2 = .O98, p < .05) , although m n e of the MMPI factors significantly discriminated the
extrafamilial offense group from the average of the
intrafamilial and chronic offense groups (multivariate R~ =
. 0 7 7 , ns).
The relationship between the unique aspects of the MCMI
factors and the contrast coded offense variables indicated
that individuals belonging to the chronic offense clusters
were significantly more likely to obtain a higher score on
the MCMI Bizarre/withdrawn factor and a lower score on MCMI
Somatoform factor than those belonging to the intrafamilial
offense cluster (multivariate R~ = .O85, p < .O5) . Individuals belonging to the extrafamilial offense cluster
were more likely to obtain a higher score on the MCMI
Paranoid factor than were the intrafamilial and chronic
groups (multivariate R~ = .082, p < .05) .
CHAPTER IV
DISCUSSION
The purpose of the present study was to determine
whether the MCMI could better differentiate sex offenders on
the basis of offender, offense/victim characteristics, than
could the MMPI. The rationale for this was that sex
offenders possibly harbour more subtle personality
differences, best captured by the MCMI, than more
pathological clinical syndromes that the MMPI was designed
to assess. An additional goal of this study was to include
more offender, 'offense, and victim characteristics than
previous MMPI studies, as these variables may confound
results across studies, and make coiparison of results
difficult.
The first part sf the analysis examined the mPI and
MCMI separately in much the same manner as in previous
studies, to facilitate comparison of the present sample with
those used in other studies of sex offenders. The means,
and modes for each scale were calculated, as were the order-
independent two-point codes. The mean and modal MMPI 2-
point code was 2-4/4-2, and was exhibited by 6.3 percent of
the total sample.
This result corroborates previous studies of sex
offenders (Groff & Hubble, 1984; Erickson et al., 1987;
Hall, 1989). Groff & Hubble (1984) did not report the
percent of their sample that obtained the 2-4/4-2 code,
thereby limiting comparison of their study with the present
research. Erickson et al. (1987) found that only 12.6
percent of their sample of child molesters produced the 2-
4/4-2 profile. Hall (1989) found that only 13.6 percent of
his sample of male sexual offenders against children
produced a 2-4/4-2 2-point code, and concluded the sample
was characterized by multiple MMPI elevations, and that
"sexual offenders are not characterized by a specific MMPI
profile configuration." (page 411).
In the present sample, although the 2-4/4-2 2-point
code was the mean and modal profile, only eight subjects, or
6.3 percent of the total subjects exhibited this profile.
This was a Lower percentage than obtained in Erickson et al.
(1987) and Hall (1989), and could perhaps partially be
explained by the fact that the present study included males
that offended against adults, as well as offenders against
children. However, it would appear that, in agreement with
Hall (1989), no specific MMPI 2-point code profile
configuration characterizes sexual offenders.
Also, a 4-2/2-4 profile sheds little insight into
specific personality characteristics of an individual
obtaining this 2-point code. Lachar (1973) suggested that
this code type is frequently exhibited by a psychopathic
personality who has been ncaughtu, and is consequently
reacting to situational, versus internal, stress.
Therefore, even if more individuals in this study obtained
such a profile, one could not elaborate on specific
personality characteristics, other than to state that
pending legal consequences appear to be precipitating
psychological discomfort.
The MCMI was analyzed in the same manner as the MMPI,
and this yielded a mean 2-point code of 3-A/A-3 (Dependent-
Anxious) for the total sample. However, the mean scores for
these scales were not clinically elevated. Unlike the MMPI,
the mean and modal 2-point MCMI code were not the same. The
modal order-independent 2-point code for the MCMI was D-A/A-
D (Dysthymic/Anxiety; N=21, 16.5%). Again, a D-A/A-D MCMI '
profile may be situation specific, in that an individual
facing a trial situation is likely to anxious, depressed,
apprehensive, and apathetic. ,
The above analysis essentially fails to illuminate the
personality characteristics of sex offenders, and a salient
MMPI or MCMI profile type for the total sample did not
emerge. Additionally, one may expect individuals to exhibit
the above MMPI/MCMI modal profiles on the basis of their
situation alone. Thus, a more refined statistical technique
was required to explore the underlying personality
differences in the present sample of sex offenders.
The above analysis was carried out on the total sample
of sex offenders, ignoring variation of offender, and
offense/victim characteristics. As previously mentioned,
this has been an inherent problem with previous MMPI
studies. However, the addition of more offender, offense,
and victim variables created a problem of equal magnitude,
namely the subject/variable ratio increases and the power of
the study decreases. In order to circumvent this problem
the number of personality variables (totalling 33 MMPI and
MCMI scales) was reduced to eight factors, four factors for
each of the MMPI and MCMI instruments. The number of
offender, offense, and victim variables was also reduced
from thirteen to three with the aid of cluster analyses.
Thus, collapsing the total number of variables from 46 to 11
served the useful purpose of allowing the inclusion of many
potentially important variables without compromising the
power of the statistical analyses. This method of analysis
represents a pivotal departure from previous MMPI research >
with sex offenders. Rather than a priori defining groups of
sex offenders as in previous research, the offense and
offender groups were defined empirically. This appears
preferable to assigning subjects to groups that are
determined by an arbitrary set of criteria for group
membership.
Cluster analysis of the offender variables yielded two
meaningful solutions, one producing two clusters, the other
producing three clusters. Cluster analysis of the offense
variables produced one meaningful solution of three
clusters. These clusters will be summarized here to
facilitate remaining discussion.
Offender Variables
(2 cluster solution)
ona abusive History: - unremarkable history in terms of physical and sexual abuse
- no previous criminal record
Abusive History: - likely to have been physically and/or sexually abused - likely to have a previous criminal record
(3 cluster solution)
Offense Variables (3 cluster solution)
Young : - 30 or younger - employed at unskilled labour - single or in 1st marriage
Unemployed: - between ages 31-40 - unemployed - equally likely to be divorced, single, or married
Married: - oldest of the 3 clusters - employed as skilled laborers, or professionals
- married
Intrafamilial: - one victim - intrafamilial offense - victim 10 years old or younger - 20 or more offense repeats - relatively more serious offense Extrafamilial: - one victim - extrafamilial offense - victim 11 years old or older - one offense - relatively less serious offense Chronic : - two or more victims - both intrafamilial and extrafamilial offenses - victims 10 years old or younger - between 2 and 19 offense repeats - relatively more serious offense
The present study was essentially atheoretical in that
there was no a priori hypotheses regarding the manner in
which the offenses would be grouped. For example, Grothls
categorization of fixated versus regressed pedophiles was
not necessarily expected to capture the offense groupings.
However, the intrafamilial offense cluster roughly
corresponds to Groth's regressed pedophile category, whereas
the chronic group approximates the fixated category, insofar
as the victims were children. The intrafamilial offenses
were most often committed by married individuals, indicating
that children were not the only preferred object of sexual
orientation. Also, these offenses, by definition, were most
often committed within the family suggesting that these
individuals were not actively searching for victims, rather
they were more situational. The chronic offenses, on the
other hand, were often both intrafamilial and extrafamilial,
indicating that these individuals were actively seeking
sexual contact with children under the age of eleven. The
extrafamilial group does not precisely fall into either of
Grothls categories, primarily because the victims were age
eleven or older, the offense itself was typically of a
relatively minor nature, and a one-time occurrence.
As discussed in the Introduction, previous studies
using the MMPI to discriminate groups of sex offenders
typically a priori categorized groups of sex offenders based
on various offense characteristics (e.g. rapists of adults
versus rapists of children) and then compared the MMPI
profiles of each group to determine if they differed
significantly. Stated another way, the offense
characteristics were considered the independent variable and
the MMPI profiles were viewed as the dependent variable. In
this study these roles were reversed such that the
personality profiles, along with offender characteristics,
were used to predict the offense types, or in this case,
offense clusters.
The interrelationships among the personality, offender,
and offense variables were investigated using set
correlation analysis. This procedure was used to determine
both the common and unique contributions of personality
variables and offender characteristics to offense variables.
The relationship between the two groups of independent ,
or predictor variables (personality and offender groups)
revealed that individuals with a history of physical and/or
sexual abuse (two cluster solution of offender variables)
were more likely to score higher, on average, on three
personality factors. These were MMPI ~mpulsive/moody and
Sensitivity factors and the MCMI Bizarre/withdrawn factor.
This is not an unexpected finding, given that individuals
with an abusive background would likely harbour more
personality disorders than individuals who did not
experience abuse during childhood. Not surprisingly, the
three cluster solution for the offender variables, which
consisted of demographic characteristics such as marital
status, employment, and age, were not significantly
correlated with the MMPI and M C M I personality factors.
The criterion variables, or offense clusters were
contrast coded before their relationship with the predictors
was investigated. The intrafamilial and chronic offenses
were relatively more serious than were the extrafamilial
offenses. They typically involved younger victims, more
than one repeat of the offense, and the offense itself
typically included oral sex, digital, vaginal, and/or anal
penetration. However, if one of these offense types could
be classified as the more serious of the two, it would be
the chronic offenses, primarily because they often involved
more than one victim, and these were both intrafamilial and
extrafamilial. The discussion to follow will describe each
of the offense clusters (intrafamilial, chronic, and
impulsive) separately, and the personality and offender
factors that predict them.
The chronic offenses were most often committed by the
Unemployed offender group, and these individuals frequently
had an abusive background. When compared with the
intrafamilial offense group, the chronic offense group
exhibited significantly higher scores, on average, on the
MCMI ~izarre/withdrawn factor and MMPI somatic factor.
Recall that the M C M I scales loading on the
~izarre/withdrawn factor were psychotic ~epression,
Avoidant, and schizoid. The Avoidant and Schizoid scales
belong to Millon's (1983) basic personality patterns (Axis
11) category. The eight scales from this category reflect
relatively pervasive and enduring traits that are indicative
of premorbid characterological functioning. According to
Millon's (1986) theoretical derivation of pathological
personalities, both scales represent a detached personality
style. This personality style reflects an inability to
experience pleasure from self or others, and a minimal sense
of contentment and joy. Detached types tend to exhibit
increasingly self-alienated and isolated behaviors.
The psychotic Depression scale, which also loads on the
~izarre/withdrawn factor, belongs to the clinical symptom
syndromes category (Axis I), reflecting a more reactive,
transient state. Typically the symptom is precipitated by ,
external events (in this case perhaps a pending trial),
which accentuate the more everyday features of the basic
personality style.
The chronic offense group also scored higher than the
intrafamilial offense group on the MMPI Somatic factor,
indicating a demanding, complaining personality style,
manipulativeness, lack of insight, hostility, repression,
denial, and attention-seeking behaviour. Relationships with
others tend to be superficial. ~epression and anxiety are
often conspicuously absent. Individuals belonging to the
chronic offense group likely do not feel guilt or remorse
about their offense, and deny responsibility for their
actions. Additionally, it is probable that these
individuals would have a poor prognosis for treatment, but
further research would be required to substantiate this '
speculation.
Intrafamilial offenses were most often committed by the
married offender group. These individuals were also likely
to have had an abusive background. When compared to the
chronic offense group, Intrafamilial offenses were more
likely, on average, to be committed by individuals with
higher scores on the M C M I Somatoform factor.
The M C M I scales that loaded on the Somatoform factor
were Somatoform, Anxiety, Dysthymic, and Borderline. The
first three scales belong to the clinical symptom syndromes
category, while the Borderline scale belongs to pathological
personality disorders category. The latter reflects a
chronic pathology in the overall structure of personality,
and typically individuals exhibiting elevations on one or
more of these scales evidence modest levels of social
competence, due to a failure to learn from previous
difficulties and the tendency to incite self-defeating
circles. High scores on the M C M I Somatoform factor indicate
somatic tension, anxiety, dysphoria, massive interpersonal
ambivalence, feelings of guilt, and a lack of antisocial
behaviour. It appears that these offenders are anxious and
likely to harbour feelings of guilt and confusion about
their offenses.
Before discussing the third offense group
(extrafamilial offenses), it would be useful at this point
to elaborate on the distinction between the MMPI Somatic
factor and the MCMI Somatoform factor. This is especially
important given that these two factors are not significantly
correlated (refer to Table 12), and each predict a different
offense type. The primary distinction between the two
factors appears to be guilt and anxiety components. Those
individuals with high scores on the MCMI Somatoform factor
likely possess a capacity to experience feelings of guilt
and remorse, and a corresponding lack of antisocial
behaviour. These individuals also appear to be highly
anxious. Individuals scoring high on the MMPI Somatic
factor, on the other hand, are characterized by a lack of ,
guilt feelings, and may harbour antisocial personality
features. Additionally, in contrast to individuals scoring
high on the MCMI Somatoform factor, those scoring high on
the MMPI Somatic factor do not present as anxious. The
common theme between the two factors is that physical
complaints tend to be psychological in origin.
The extrafamilial offenses were most likely to be
committed by the young offender group. These individuals
seldomly had an abusive background. Recall that the
impulsive offenses were considered the least serious of the
three offense groups, and were compared with the average of
the intrafamilial and chronic offenses.
The Extrafamilial offenses were consistently
discriminated from the Chronic and Intrafamilial offenses by
the MCMI Paranoid factor. The Paranoid and psychotic
elusions scales loaded on the MCMI Paranoid factor,
indicating paranoid behaviour and thinking with the
associated features of suspiciousness, mistrust,
defensiveness, persistent ideas of self-importance and self-
reference, and hostile acting out. The Paranoid and
Psychotic Delusions scales fall into the clinical symptom
syndromes and pathological personality disorders categories
respectively.
The following is a summary of the personality factors
and offender characteristics that predict each of the
offense groups: ,
MCMI Bizarre/withdrawn, MMPI somatic factors; Unemployed
offender group - Chronic offenses MCMI Somatoform factor; ~arried offender group - Intrafamilial offenses
MCMI Paranoid factor; Unskilled Labour group - Extrafamilial offenses
In general, it appears that individuals who are between the
ages of 31-40, unemployed, have been physically and/or
sexually abused, exhibit a detached personality style, and
do not appear to be anxious, depressed, or remorseful, tend
to commit relatively serious, chronic offenses. These
offenses are likely to be both intrafamilial and
extrafamilial, against children under the age of eleven, and
tend to include oral sex, digital, vaginal, and/or anal
penetration.
Individuals who are employed as professionals or
skilled labourers, married, have been physically and/or
sexually abused, exhibit anxiety, dysphoria, feelings of
guilt, and a lack of antisocial behavior, tend to commit
relatively serious, intrafamilial offenses. These offenses
are typically committed numerous times against one victim
under the age of eleven, and also included oral sex,
digital, vaginal, and/or anal penetration.
Finally, individuals who are employed at unskilled
labour, are under the age of 30, single, have not been
physically or sexually abused, and are suspicious,
defensive, and mistrustful typically commit Extrafamilial
offenses. These offenses are less serious in nature than
the previous two in that they are Ithands offN (e.g. obscene
phone calls, exhibitionism), or fondling only, and the
victims are age eleven or older. ~dditionally, there is
typically one victim only, the offense is extrafamilial, and
a one-time occurrence.
The original question that this study set out to
explore was whether the MCMI would better discriminate sex
offenders on the basis of offender, offense, and victim
characteristics, than would the MMPI. It appears from the
results discussed above that the MCMI consistently
discriminated the Chronic from the Intrafamilial offenses,
as well as the Extrafamilial offenses from the average of
the Chronic and Intrafamilial offenses. The MMPI only
discriminated the chronic and intrafamilial offenses, and
did not discriminate between the less serious extrafamilial
offense group and the more serious chronic and intrafamilial
groups.
Therefore, it appears that the MMPI is useful in
discriminating sex offense types to the extent that the
offenses are relatively serious, and enduring in nature.
The MCMI, on the other hand, indicated greater predictive
validity in the present study, in that it discriminated the
more serious and less serious offense types. Thus, based on
the results of the present study, both personality >
instruments are useful in the assessment of sex offenders,
but if given a choice between the two, the MCMI has
demonstrated superior predictive validity.
Limitations of Current Study and Sussestions for Future
Research
The current findings are compromised by the following
limitations:
1. The current study did not include a control group for
comparison of MMPI and MCMI protocols. It would be useful
to determine whether MMPI and MCMI protocols differed
between sex offenders and ronsex offenders, and between sex
offenders and normal, nonclinical populations.
2. It would be interesting to determine whether the factor
structures of the MMPI and MCMI would be replicated in
another sample of individuals charged with sexual offenses.
3. There were several variables that were restricted in
range due to sample size, and were deleted from analyses.
For example, most of the offenders were heterosexual,
offended against females, were Caucasian, and raised by
their biological father and mother. A larger sample size
might have resulted in greater variability or at 1east.more
power to handle such restricted variance, and may provide
interesting results.
4 . The results to the present study provide some insight
into the underlying personality characteristics of a ,
socially deviant population. Unfortunately, the results
have limited utility for aiding the practising clinician in
individual assessments. However, if the results of the
present study were replicated, a new scale could be derived
that would be useful in discr'iminating various types of sex
offenders.
5 . Implications for treatment and recidivism based on
personality characteristics were not explored and clearly
would be a relevant topic for future research. The results
of the present study could not be used prospectively to
predict individuals who would be likely to commit a sexual
offense because subjects were selected for research after
committing an offense and were not compared to a random
sample of normal controls. However, the current results nay
be have implications for the issues of treatment and
recidivism. The intrafamilial and extrafamilial offender
groups showed personality characteristics that indicate
amenability to treatment. The intrafamilial group typically
harboured guilt feelings that could be used in a therapeutic
manner and prevent further offending. ~ikewise, the
extrafamilial group typically had offended only once and
treatment would likely be successful in preventing
recidivism. The chronic offender group, on the other hand,
exhibited antisocial tendencies, and may be more resistant
to therapy. Therefore, this group may be the most likely of
the three to reoffend. ,
CHAPTER V
REFERENCES
American Psychiatric Association (1987). Diasnostic and . Statistical Manual of Mental Disorders DSM-111-R. Washington.
Anderson, W.P., Kunce, J.T., & Rich, B. (1979). Sex offenders: Three personality types. Journal of Clinical Psvcholosv, 3 5 ( 3 ) , 671-676.
c Armentrout, J.A., & Hauer, A.L. (1978). MMPI1s of rapists ,
of adults, rapists of children, and non-rapist sex offenders. Journal of clinical Psycholosy, 34(2), 330- 332.
#Butcher, J.H. & Tellegen, A. (1978). Common methodological problems in MMPI research. Journal of Consultins clinical Psvchol~~~, 46(4), 620-628.
Carroll, J.S. & Fuller, G.B. (1971). An MMPI comparison of three groups of criminals. Journal of clinical Psvcholosv, 2 7 , 240-242.
Cohen, J. (1982). Set correlation as a general multivariate data-analytic method. Multivariate Behavioral Research, 17, 301-341. -
Erickson, W.D., Luxenberg, M.G., Walbek, N.H., & Seely, R.K. (1987). Frequency of MMPI two-point.code types among sex offenders. Journal of consultins and clinical Psvcholoc~y, 55(4) , 566-570.
Finkelhor, D., & Araji, S. (1986). Explanations of pedophilia: A four factor model. Journal of Sex Research, 22(2), 145-161.
Gearing, M.I. (1979). The MMPI as a primary differentiator and predictor of behaviour in prison: A methodological critique and review of the recent literature. ~svcholo~ical Bulletin, El6(5), 929-963.
Goldberg, L.R. (1972). Man versus mean: The exploitation of group profiles for the construction of diagnostic classification systems. Journal of Abnormal Psvcholosv, - 79 121-131. - I
Greene, R. (1980). The MMPI An Interpretive Manual. New York: Grune & Stratton.
Greene, R. (1978). An empirically derived MMPI carelessness scale. Journal of clinical Psvcholosy, - 34, 407-410.
Greer, S.E. (1984). A review of the Millon clinical Multiaxial Inventory. Journal of Counselins and Development, 63, 262-263.
Groff, M.G. & Hubble, L.M. (1984). A comparison of father- daughter and stepfather-stepdaughter incest. criminal Justice and Behaviour, ll(4), 461-475.
Hall, G.C., Maiuro, R.D., ~italiano, P.P., & Proctor, W.C. (1986). The utility of the MMPI with men who have sexually assaulted children. Journal of consultins and Clinical Psvcholosv, 54(4), 493-496.
Hall, G.C. (1989). WAIS-R and MMPI profiles of men who have sexually assaulted children: Evidence of limited utility. Journal of Personality Assessment, 53(2), 404- 412.
Hartmann, B.J. (1967). comparisons of selected experimental MMPI profiles of sexual deviates and sociopaths without sexual deviation. Psvcholosical Reports, 20, 234.
Lachar, D. (3.973)~ The MMPI: clinical Assessment and Automated Interpretation. Los Angeles, CA: Western Psychological Services.
Langevin, R., Paitich, D., Freeman, R:, Mann, K., 81 Handy, L. (1978). Personality characteristics and sexual anomalies in males. Canadian Journal of Behavioral Sciences, u(3), 222-238.
LaTorre, R. (1987). Sex offender assessment and treatment proqram. Unpublished manuscript.
Marsh, J.T., Hilliard, J., & Liechti, R. (1955). A sexual deviation scale for the MMPI. Journal of Consultins Psvcholosv, m(1) , 55-59.
McCormack, J., Barnett, R., & Wallbrown, F. (1989). Factor structure of the Millon Clinical Multiaxial Inventory (MCMI) with an offender sample. Journal of Personalitv Assessment, 53, 442-448.
McCreary, C.P. (1975a). ~ersrnality differences among child molesters. Journal of Personality Assessment, 39(6), 591- 593.
McCreary, C.P. (1975b). personality profiles of persons convicted of indecent exposure. Journal of Clinical Psvcholoqy, 31, 260-262.
Millon, T. (1969). Modern Ps~chopatholocrv: A Biosocial A~proach To Maladaptive Learnincr and Functioninq. ~hiladelphia, PA: Saunders.
Millon, T. (1977). Millon Clinical Multiaxial Inventory Manual. National Computer Systems, Inc., Minneapolis, MN .
Millon, T. (1981). Disorders of personality: DSM-111. Axis 11. New York: Wiley. -
Millon, T. (1985a). The MCMI and DSM-111: Further commentaries. Journal of ~ersonalitv Assessment, 50(2), 205-207.
Millon, T. (1985b). The MCMI provides a good assessment of DSM-I11 disorders: The MCMI-I1 will prove even better. Journal of Personality Assessment, 4 9 ( 4 ) , 379-391.
Millon, T. (1986). The MCMI and DSM-111: Further commentaries. Journal of Personality Assessment, 50, 205- 207. 5
Panton, J.H. (1978). Personality differences appearing between rapists of adults, rapists of children, and non- violent sexual molesters of female children. Research Communications in Psycholoqy, Psychiatry and Behaviour, 3(4), 385-393.
Panton J.H. (1979). MMPI profile configurations associated with incestuous and non-incestuous child molesting. Psvcholouical Reports, 45, 335-338.
Peek, R.M. & Storms, L.H. (1956). Validity of the Marsh- Hilliard-Liechti MMPI Sexual Deviation scale in a state hospital population. Journal of Consultins Ps~choloqy, = ( 2 ) , 133-136.
Perdue, W. & Lester, D. (1972). Personality characteristics of rapists. Perceptual and Motor Skills, 3 5 , 514.
Persons, R.W. & Marks, P.A. (1971) . The violent 4-3 MMPI personality type. Journal of Consulting and Clinical
Quinsey, V.L., Arnold, L.S., & Pruesse, M.G. (1980). MMPI profiles of men referred for a pretrial psychiatric assessment as a function offense type. Journal of Clinical Psycholoqy, 36(2) , 410-417.
Rader, C.M. (1977). MMPI profile types of exposers, rapists, and assaulters in a court services population. Journal of Consultins and Clinical Psvcholoqy, 45(l), 61-69.
Scott, R.L. y& Stone, D.A. (1986). MMPI profile constellations in incest families. Journal of Consultinq and Clinical Psycholoq~, 54(3), 364-368.
Swenson, W.M. & Grimes, B.P. (1958). characteristics of sex offenders admitted to a Minnesota state hospital for pre- sentence psychiatric investigation. psychiatric Quarterly Sup~lement, 3l, 110-123.
Toobert, S., Bartelme, K.F., & Jones, E.S. (1959). Some factors related to pedophilia. International Journal of Social Psvchiatrv, $, 272-279.
Wattron, J.B. (1958). Validity of the ~arsh-~illiard-Liechti MMPI Sexual Deviation scale in a state prison population. Journal of Consultinq Psvcholoqy, 22(l), 16.
Widiger, T.A. & Sanderson, C. (1987). The convergent and discriminant validity of the MCMI as a measure of DSM-I11 personality disorders. Journal of ~ersonalitv Assessment, 5 l , (2), 228-242.
Widiger, T.A., Williams, J.B., Spitzer, R.L., & Francis, A. (1985). The MCMI as a measure of DSM-111. Journal of Personality Assessment, 4 9 ( 4 ) 366-378.
Widiger, T.A., Williams, J.B., Spitzer, R.L., & Francis, A. (1986). The MCMI and DSM-111: A brief rejoinder to Millon (1985). Journal of Personality Assessment, 50(2), 198- 204.