Prediction of incidents in a forensic psychiatric facility using demographic and psychological test variables and identi...


Material Information

Prediction of incidents in a forensic psychiatric facility using demographic and psychological test variables and identification of naturally occurring subgroups of forensic inpatients using cluster analysis of Minnesota Multiphasic Personality Inventory profiles
Physical Description:
viii, 257 leaves : ill. ; 29 cm.
Bordini, Ernest John
Publication Date:


Subjects / Keywords:
Forensic Psychiatry -- methods   ( mesh )
Hospitals, Psychiatric   ( mesh )
MMPI   ( mesh )
Clinical and Health Psychology thesis Ph.D   ( mesh )
Dissertations, Academic -- Clinical and Health Psychology -- UF   ( mesh )
bibliography   ( marcgt )
non-fiction   ( marcgt )


Thesis (Ph.D.)--University of Florida, 1988.
Bibliography: leaves 248-256.
Statement of Responsibility:
by Ernest John Bordini.
General Note:
General Note:

Record Information

Source Institution:
University of Florida
Rights Management:
All applicable rights reserved by the source institution and holding location.
Resource Identifier:
aleph - 000983935
oclc - 20302911
notis - AEW0102
System ID:

This item is only available as the following downloads:

Full Text







Copyright 1988


Ernest John Bordini

Al caro Zio Gianni

con ammirazione, e gratitudine.


This research involved a three year odysey which would not have

been possible without much cooperation and support. My first

thanks is to Dr. Jacquelin Goldman who supported this research

from its beginnings and whose enthusiasm, teaching, and clinical

experience served as inspiration. The support and constructive

criticism of Dr. Davis, Dr. Bauer, Dr. Glaros, and the perspective of

Dr. Von Mering contributed much to this undertaking.

Research in institutions requires much cooperation from the

staff and administration of the institution. Much gratitude is

due to the staff and administration of NFETC. The staff's pride,

interest, and pursuit of knowledge is evident in their support of

research and in the quality of care delivered.

Finally, such an odysey would likely shipwreck without the

necessary emotional support through trying times and obstacles.

For this, I am forever grateful to my Marianne.



ACKNOWLEDGEMENTS ................................. ...... iii

ABSTRACT ................................................. vi


ONE INTRODUCTION..................................... 1

Overview of the Literature on Prediction of
Violence ..................................... 1
Historical Background ........................... 3
Early Research on the Prediction of
Dangerousness ................................. 7
Longitudinal Research of the Correlates of
Violent Behavior ............................ 14
Psychological Test Correlates of Violent
Behavior ....................................... 17
Cluster Analytic Approaches to Classification.... 26
Studies of Predictors of Institutional
Aggression ................................... 28
Prediction Issues ............................. 31
Statement of the Problem ........................ 34

TWO METHOD........................................... 38

Subjects ......................................... 38
The Institution............................... 39
Materials and Measures.......................... 40
Dependent Measures.............................. 49
Analyses ................ ........ ................. 52

THREE RESULTS .......................................... 61

Sample Characteristics.......................... 61
Comparison of Residents Tested by MMPI and
Rorschach............................ .......... 75
Prediction Equations................. .......... 85
Cluster Analyses ................................. 145
Comparisons of Residents Involved and Not
Involved in Incidents.................. ...... 161


FOUR DISCUSSION....................................... 206

Methodological Considerations ................... 206
Demographic Data ................................ 210
Test Characteristics............................ 212
Incidents ........................................ 216
Blockwise Multiple Regression Analyses .......... 216
Summary of Prediction Hypotheses ................ 223
Post-Hoc Analyses ................... .......... 235
Conclusions ...................................... 240


INCIDENT/USE OF FORCE REPORT .................... 245

REFERENCES................................ .... .. ........... 248

BIOGRAPHICAL SKETCH........................................ 257

Abstract of Dissertation Presented to the Graduate School
of the University of Florida in Partial Fufillment of the
Requirements for the Degree of Doctor of Philosophy



Ernest John Bordini

April 1988

Chairperson: Dr. Jacquelin Goldman
Major Department: Clinical and Health Psychology

Psychologists and psychiatrists often render opinions regarding

the potential dangerousness of individuals in the context of forensic

evaluations and involuntary commitment proceedings. However, early

research suggested predictions of dangerousness resulted in high false

positive rates. In 1984, it was suggested the early research was

overgeneralized and suggested using multivariate and actuarial methods

in more acutely disturbed populations.

The present research investigated the ability of multiple

regression analyses of demographic and psychological test data to

identify residents with high base rates of involvement in institutional

and violent incidents in a maximum security forensic facility. Data

were collected for 451 male residents, the majority of which were

incompetent to stand trial.

A blockwise stepwise multiple regression procedure was used to

develop equations predictive of total incident rate, two types of

aggressive incident rates, use of force incident rates, and fighting

incident rates. Regression models identified groups of residents with

a 75% base rate of incidents, and a slightly greater than 50% base rate

of violent incidents.

Cluster analysis of Minnesota Multiphasic Personality Inventory

(MMPI) profiles has been a useful methodology to identify subgroups of

criminal offenders. The present investigation attempted to identify

naturally occurring subgroups of forensic inpatients by cluster

analyzing 188 MMPI profiles. Results of the analyses indicated a

possible six cluster and a possible three cluster solution. Analyses

failed to validate the six cluster solution.

Significant between group differences were found for the three

cluster solution. The group MMPI profiles differed in terms of

overall profile elevation. The highest MMPI profile elevation

group had the longest mean length of treatment, greatest

proportion of residents involved in use of force incidents, and

greatest proportion of residents with substance abuse histories.

The present study also examined between group differences for

residents involved and not involved in incidents. Differences

with respect to race, education, diagnosis, and test results were

found and discussed in the context of previous research.




Overview of the Literature on Prediction of Violence

In recent years, research on the prediction of violence has been

sparse. Research in the area of forensic psychology shifted

away from an early focus on the prediction of violence, to the

rights of involuntarily committed patients, and later to the

characteristics of insanity acquitees. These shifts often

paralleled political and social concerns of the times. Research

in the area of prediction of violent behavior has been relatively

neglected despite the American Psychiatric Association's (1977)

description of this as one of the most important areas for

research in the interface of psychiatry and the law.

Monahan's (1981) frequently cited book, Predicting Violent

Behavior, may have had an adverse impact on research efforts to

attempt such predictions. His review, which included a focus on

several large scale studies of chronically institutionalized

patients, and studies with overall low incidence rates of

violence, was grossly overinterpreted as indicating that

predictions of violence could not be made, or at most are only

accurate one out of three times. Monahan (1984) concurred that


his review had been overgeneralized by other professionals and

made an appeal for renewed efforts in the area.

Monahan (1984) as well as others (Shah, 1978) suggested the use

of multivariate models which focus on short term predictions of

violent behavior within populations known to have high base rates

or histories of violence. Meehl (1954, 1986) has long advocated the

development of statistical models of prediction. Meehl argued

that such models outperformed clinical predictions and were more

explicit and testable.

Renewed research interest in prediction of violent behavior

has been stimulated by recent events. The increase in the number

of mentally ill involved in the criminal justice system, the

notoriety of Hinkley's assassination attempt on President Reagan,

and the Tarasoff decision have captured media attention with

respect to violent behavior by the mentally ill and has renewed

interest in determinations of dangerousness.

The present study attempted to identify a group of individuals

with a greater than average frequency of institutional violence

by applying a multivariate multiple regression model. The use of

a stepwise regression model allowed for reduction of the number

of possible predictor variables by selecting only those variables

at each stage which added significantly to the variance accounted

for. This technique is suited for research that is primarily

predictive or exploratory in purpose.

A second goal of the present research was to provide additional

information on the characteristics of a population of forensic


inpatients which consisted mostly of individuals adjudicated

incompetent to stand trial. This group has been somewhat

neglected in forensic research. Identification of subgroups of

forensic inpatients was attempted by a cluster analysis of Minnesota

Multiphasic Personality Inventory (MMPI) profiles. This methodology was

previously successful in identifying groups of criminal offenders

(Megargee & Bohn, 1979).

Historical Background

Research conducted in the interface of psychiatry and the law

has shifted in focus over time. Steadman (1984) noted that much of

the earliest research conducted in the early 1960s focused on

the definition of insanity and the implementation of the insanity

defense. Research found that variations of the insanity plea such

as the M'Naughton rule, the ALI test, mens rea, or the Durham

rule had little effect in the number of people who were found not

guilty by reason of insanity (NGBRI). Insanity acquittals were

found to be successful in approximately 2% of the cases in which

insanity was raised as a defense regardless of the particular test

which was applied (Johnson, 1975).

Research conducted in the late 60s shifted in focus to the civil

rights issues of those persons involuntarily hospitalized. This

led to research examining forensic and civil commitment

proceedings and the assumed dangerousness of the committed

forensic patient. Much of this research led to the release of

many individuals who had been involuntarily hospitalized for many


years. Monahan's (1981) review which included studies of these

released patients suggested that predictions of dangerousness

resulted in error at least two out of three times.

Much of the research in the 70s examined the social,

demographic and psychological characteristics of insanity

acquitees. Steadman (1984) noted that an inverse relationship

exists between the amount of research conducted and the size of

the forensic population. Insanity acquitees which comprise

approximately 8% of the mentally ill forensic population have

been the most studied, while mentally disordered sex offenders,

those adjudicated incompetent to stand trial, and the mentally

ill inmate have been largely ignored.

Research in the 70s and early 80s has neglected the prediction

issue despite the fact that since 1970 most states have switched

the requirements for involuntary commitment from one that focused

on need for treatment to one that uses dangerousness as a

criterion (Monahan, 1984). Steadman (1984) emphasized that

there is a large gap between what has been thought to be securely

documented in the area of prediction of dangerousness and what

evidence there exists.

Despite pessimism about the accuracy of predictions of violence

such predictions have been a historical part of the legal

process. Determinations of dangerousness are often embedded in

prognostic statements, classifications, and placements within the

correctional system (Shah, 1978).


Clinicians concerns about the potential dangerousness of the

clients they are treating have been raised by the litigious

medico-legal climate. The Tarasoff decision (Tarasoff vs Regents, the

University of California, 1976), which is frequently overinterpreted as

creating a clinician's "duty to warn" potential victims, made the

prediction issues more salient. Although the same court revised

the language to read that a more general duty to exercise

reasonable care exosts (Givelber, Bowers, & Blitch, 1985), the

controversy concerning clinicians' ability to predict violent

behavior and their responsibility with regards to potential

victims spread to other states and raised issues concerning possible

malpractice suits (Ridgewood Financial Institute, 1985).

The increase in the rate of psychiatric patients with histories

of arrest or violent behavior has renewed research interest with

respect to the dangerousness of the mentally ill. The number of

the mentally ill who are involved in the criminal justice system

has increased. Monahan (1981) reported that in 1947 fifteen

percent of New York patient releases had prior arrests compared

to 40% in 1975. A similar trend is evident in England. Hinton

(1983) reported that in English "security hospitals" there was a

75% increase in admission rates for psychopathic disorders

between 1961 and 1970.

Perhaps as a consequence of deinstitutionalization, more mentally

ill individuals are being arrested. Research conducted after the

1950s indicates that discharged psychiatric patients had higher


arrest rates than the general population whereas research prior

to that time indicated the reverse was true (Monahan, 1981).

Sosowsky (1978) reported that patients admitted to a psychiatric

hospital without a criminal record had subsequent arrest rates

three times that of the general population, and were arrested

five times as often as the general population for violent crimes.

Rubin (1972) reviewed the literature available at that time and

concluded that the reversal in the arrest rates of the mentally

ill was attributable to the increased coexistence of antisocial

behavior and mental illness for young, poor, unemployed, and

unskilled males. Teplin (1984) reported that for similar

offenses mentally disordered citizens are more likely to be

arrested than other citizens and suggested that the trend toward

deinstitutionalization was a factor in this process.

Some of the neglect with respect to research involving prediction

of violence may be a result of the overgeneralization from early

research that clinicians had little ability or expertise in

making such predictions. The five studies which Monahan (1984)

described as forming the core of "first generation" research on

predictions of violence examined predictions that those released

from long-term custodial institutions would engage in violent

behavior. The disappointing results of these attempts were most

likely attributable to the low base rates of the target behavior,

treatment effects, aging, and unspecified effects of long term



Early Research on the Prediction of Dangerousness

Monahan (1984), whose review of the efficacy of clinical

predictions of dangerousness led to much of the pessimism about

attempts at prediction, criticized the overgeneralization and

uncritical acceptance of this "first-generation" of research.

This research suffers from flaws such as attempting to predict

rare events, studying borderline cases, generalizing

pre-hospitalization predictions to post-treatment and release

behavior, and in some studies did not involve any actual

prediction by clinicians at all.

Ethical considerations prohibit the release of individuals

everyone agrees to be dangerous. This constraint necessarily

limits investigations of post-release violence to examining the

behavior of those individuals for whom there was some

disagreement about. This most likely yields biased underestimates of

the efficacy of prediction. A criticism of some studies is that

predictions made concerning need for immediate hospitalization due to

dangerousness were overgeneralized and applied to post-treatment and

post-release behavior.

Since the studies reviewed by Monahan have been described as

forming the core of the early research, and the cause for the

overgeneralization that violence cannot be predicted, they are

briefly reviewed here.

Kozol, Boucher, and Garofolo (1972) conducted a five year

follow up study of 592 male offenders. Most were convicted of sex

crimes. Of these offenders, 226 were initially judged by the


psychiatric staff to be dangerous and were subsequently committed

to treatment. The court eventually released 49 offenders

contrary to the prediction of dangerousness by the psychiatric

staff. During the five year follow up period only 8% of those

released without a prediction of dangerousness committed a

serious assaultive offense compared to 35% of those released by

the court against the psychiatrist's advice.

The rate of false positives in the Kozol et al. study has

sometimes been cited as evidence of clinicians inability to

predict violent behavior. However, the predictions yielded a

group of individuals which were four to five times as likely to

commit a serious offense than the overall base rate. A further

criticism of this study was that it yielded a biased

underestimate of the efficacy of clinical predictions since the

identified dangerous group consisted of "borderline" cases. The

"dangerous" group actually consisted of only the 20% of cases predicted

dangerous by the clinicians for which the court disagreed (Litwack,


Another set of frequently cited studies followed up on the

court-ordered release of large numbers of institutionalized

individuals presumed to be dangerous. These studies were more

reflective of global or political predictions rather than actual

individually made clinical predictions.

The first of these studies eventually led to the release of

other chronically institutionalized patients which were being

hospitalized for indefinite periods of time because of


"dangerousness" and led to legal reforms of the laws which had

allowed this practice to exist. In this first study one thousand

institutionalized patients held past expiration of their

sentences without a hearing were transferred to civil hospitals

by court order. Steadman and Cocozza (1974) reported the

follow-up of the "Baxtrom" patients.

At the time of transfer, the average age of the Baxtrom

patients was 47. These patients had a mean length of 15 years of

continuous hospitalization. Twenty percent of these patients were

assaultive at any time during the next four years of civil commitment.

Those eventually released were followed for two and one half years.

During this period of freedom only 8% were convicted of any


A "Legal Dangerousness Scale" was constructed in an attempt

to predict subsequent violent arrests. The scale was based on the

presence of a juvenile record, number of previous arrests,

presence of prior convictions for violent crimes, and severity of

the admitting offense. Although only one of three patients

identified by the scale as dangerous were eventually rearrested

for a violent crime, this was four times the overall rearrest rate

for the Baxtrom patients.

In the Baxtrom study, Steadman and Cocozza (1974) found that

few patients over 50-years-old were rearrested for a violent

crime. Seventeen of the twenty arrested for violent crimes after

their release were less than 50-years-old and had a score

greater than five on the above Legal Dangerousness scale.


In a study of Pennsylvania prisoners who won suit for

release partially as a consequence of the Baxtrom decision,

Thornberry and Jacoby (1979) found that 14% of 438 long-term

institutionalized patients engaged in behavior injurious to other

persons within 4 years after their release. Similar to the

Baxtrom patients, the "Dixon" patients had a mean age of 47

years and were institutionalized for a mean of 14 years.

Approximately one of four of the 418 Dixon patients eventually

released were rearrested in a median release time of 30 months.

Fourteen percent of the released Dixon patients were rearrested

or rehospitalized for violent behavior.

As in findings reported from the Baxtrom study, the younger

Dixon patients were more likely to be arrested. Forty percent of

the patients younger than 35-five-years old were rearrested.

Nearly one in four of these younger patients were rearrested for

violent offenses.

A major criticism of the Baxtrom and Dixon studies is that

they did not directly address the efficacy of clinical prediction

of violent behavior. These studies did not attempt individual

determinations of dangerousness; instead they challenged the

public opinion that these patients as a group were a threat to

society. In fact, the facility which housed the Dixon patients

employed only one psychiatrist and no Ph.D. psychologists

(Thornberry & Jacoby, 1979). Litwack (1985) suggested these

studies are more appropriately described as studies of "political

prediction" than as attempts at clinical prediction.


In a more successful attempt at identifying a group of

individuals committing post-release offenses, the State of

Maryland (1978) published data for 421 patients treated for 3

years or more at the Patuxent Institution. The court released

286 patients despite a determination of dangerousness by the

psychiatric staff. Of those patients released directly from the

hospital, 46% showed a new offense on their FBI rapsheet within

three years after release. Thirty-nine percent of those

"conditionally released" indicated an offense within the three

year period. Only 7% of the 135 individuals the staff found "safe"

showed such an offense.

The results of the state of Maryland study were challenged by

conflicting data which were reported by Steadman (1977). Steadman

(1984) reanalyzed the data and found that 31% of the patients

recommended for release were arrested for a violent crime in

comparison with 41% of those predicted to be dangerous, making

the original data equivocal. The state of Maryland study can also

be criticized since the dangerous group was likely to be a biased

sample of all individuals predicted to be dangerous by the

clinicians because the group consisted of only the subset of

individuals for which the court disagreed.

In a study of institutional and post-release violence, a group

of 257 patients adjudicated incompetent to stand trial in New

York between 1971 and 1972 were followed by Cocozza and Steadman

(1976). Patients were examined by two pyschiatrists whose initial

finding of dangerousness permitted placement in a facility


administered by the Department of Correctional Services and a

finding of nondangerousness permitted placement to a civil

psychiatric hospital. The average age of this sample was

31 years.

Of these individuals adjudicated incompetent to stand trial,

60% were predicted dangerous and 40% were predicted not to be

dangerous. The judge disagreed with these predictions in 34 cases,

adjudicating as dangerous 26 the psychiatrists had not predicted to be

dangerous. These disagreements were not discarded in the reported

findings. Subjects were followed in the hospital and in the community

for three years if they were released.

The adjudications of dangerousness were more successful at

identifying individuals likely to be violent in the hospital than those

which would eventually be rearrested for a violent crime.

Those predicted or adjudicated to be dangerous were slightly more

likely to be assaultive in the hospital than those not

adjudicated as dangerous (42% versus 36%).

Ninety-six of of the 154 adjudicated dangerous by the court and

70 of the 103 adjudicated nondangerous were eventually released.

Forty-nine percent of those identified as dangerous versus 54% of those

not identified as dangerous were rearrested. Fourteen percent of the

"dangerous" individuals were rearrested for a violent offense in

comparison to 16% of those adjudicated not to be dangerous, indicating

little difference in the rearrest rates of these two groups.


Although Cocozza and Steadman (1976) interpreted their study as

clearly indicating no psychiatric expertise in predicting who

will be dangerous existed, Litwack (1985) criticized the study as

a general study of clinical prediction of violent behavior. He

indicated that the clinician's predictions were that individuals

were of immediate danger to themselves or others if left at

liberty and were not predictions of the likelihood of

institutional violence or the eventual risk of violence after

they were treated and released.

The inclusion of individuals judged not to be dangerous by the

clinicians but adjudicated so by the court as dangerous is

another criticism of the Cocozza and Steadman study as a pure

study of clinical prediction. The results are biased by inclusion

of residents predicted nondangerous by the clinicians in the

dangerous group.

This first generation of research suggested that the global

prediction that chronically institutionalized individuals are all

dangerous is unwarranted. These studies also indicate that while

clinicians may be able to identify subsamples of individuals much

more likely to exhibit violent behavior than the base rate, low

base rates of violent behavior in the population result in a

large number of false positives. Certain demographics such as

age, presence of a juvenile record, number of previous arrests,

presence of prior convictions, and severity of admitting offense

were correlates of violent behavior in these studies.


Longitudinal Research of the Correlates of Violent Behavior

While research concerning the clinical prediction of violent

behavior has been sparse, there is a substantial literature which

has researched the characteristics and histories of violent

individuals or groups of individuals by following large groups of

individuals over time. This research has generally found that

number of previous arrests, history of past violent behavior,

race, and socioeconomic status are correlated with subsequent

violent arrests.

Studies examining the role of early childhood predictors of

violence place importance on the presence of fighting in

childhood and juvenile arrest records. Justice, Justice, and

Kraft (1974) reviewed 1500 references of violence in the

psychiatric literature and found that childhood fighting, temper

tantrums, problems in school, and inability to get along with

others were behaviors reported to correlate with later violence.

The association between early violent behavior and later

aggression was also supported in a longitudinal study of 400

males from age 8 to 19. In this sample Lefkowitz, Eron,

Walder, and Heusmann (1977) reported that aggression at age 8

was the single best predictor of aggression at age 19.
In a large study, the records of all boys born in

Philadelphia in 1945 and still living there between their 10th

and 18th birthday were examined by Wolfgang, Figlio, and

Sellin (1972). Thirty-five percent of the boys had at least

one reported contact with the police by age 18. Race and


socioeconomic status were most predictive of delinquency. Thirty

percent of white males compared to 50% of nonwhite males had such

contact. Twenty-six percent of the high socioeconomic status

boys compared to 45% of the low socioeconomic status boys had

such a contact. Wolfgang (1977) reported a follow-up study which

found that by age 30 only 5% of the sample was arrested only

as adults. Individuals with a juvenile record were four times

more likely to be arrested as adults than those without a

juvenile record.

A strong relationship between the number of previous arrests

and subsequent violent behavior was evident in the above study.

One of three of those with at least one reported contact with

police by age 18 was arrested as an adult by age 30. The probability

of being arrested a fifth time given four prior arrests was 0.90.

The probability of being arrested for an FBI index offense given four

prior arrests was 0.36. Given 10 prior arrests there was a 0.42

probability of an FBI index arrest. Similar findings were found in a

Washington, DC, research project which analyzed the arrest records of

45,000 defendants. In this study, the probability of rearrest with

five or more previous arrests approached almost certainty (Shah,


Not all criminal groups have high rates of violent

rearrests. Monahan (1981) reported that an "Assaultive Risk

Screening Sheet" was used by the Michigan Department of

Corrections in 1978 to predict arrest for a new violent crime. The

study analyzed 350 variables for 2200 males released on


parole in 1971 over a 14 month period of time. A small

subgroup (5%) of individuals with a 40% recidivism rate was

identified by checking type of crime, nature of institutional

behavior, and presence of arrest before age 15. The overall

base rate of violent arrests was 10% in the Michigan study.

Murphy (1980) replicated this study and found a 32% recidivism

rate for the identified high risk group.

A series of three studies using 4,146 California Youth

Authority wards conducted by the California Department of

Correction was reported by Wenk, Robison, and Smith (1972). Only

6% of the youths studied had been committed due to a violent offense.

A scale was developed which identified a small subgroup 14% of which

subsequently committed a violent act while on parole. Only 5% of the

nonidentified group committed such an act. The use of the scale

resulted in an 86% false positive rate.

Twenty percent of the Youth Authority parolees were assigned

to a potentially aggressive category in a second study (Wenk,

Robison, and Smith, 1972). The rate of conviction and

imprisonment for this group during a one year follow up was only

three per thousand. Finally, 4000 of the youth authority wards

were followed for 15 months after their release. Prediction

of violence based on prior crime as well as 100 other variables

failed to yield better than a 95% false positive rate. The poor

results of the Youth Authority studies illustrate the difficulty

predicting low base rate events accurately.


In the longitudinal studies above fairly consistent results were

found. Presence of a juvenile record, number and type of previous

arrests, race, socioeconomic status, and childhood behavioral

difficulties were associated with later arrests and violent


Psychological Test Correlates of Violent Behavior

Psychological and demographic variables have been used in a

number of studies to discriminate between offenders who have

committed various types of crimes on the assumption that these

variables would also predict to future behavior. While most

studies examined group differences in present test results, some

studies have examined the predictive validity of various

psychological tests. The Rorschach, a projective technique, and

the MMPI, an objective test are frequently used tests which have

generated considerable research interest.


The Rorschach is a projective technique, first introduced in 1921

(Rorschach, 1921), in which the subject is asked to report what the

subject sees when presented each of 10 ambiguous ink blots. Exner

(1974) reviewed a variety of scoring systems (e.g., Beck, Beck, Levitt,

& Molish, 1961; Hertz, 1951; Klopfer, Ainsworth, Klopfer, & Holt, 1954;

Piatrowski, 1957; and Rapaport, Gil, & Schaefer, 1946) and, borrowing

freely from these, developed the Comprehensive Rorschach System (Exner,

1974). Exner's system provided a more standardized method which led to


renewed research interest and instruction in the use of the test

(Hertz, 1987; Ritzler & Alter, 1986; and Piotrowski, Sherry, & Keller,


The Rorschach has been frequently used to aid in the

discrimination between violent and nonviolent persons (e.g., Finney,

1955; Rose & Bitter, 1980; Shagoury, 1971). A number of Rorschach

indices, described below, have been correlated with anger, aggression,

and psychopathology.

Rorschach R is defined as the number of responses to the

cards. It is generally interpreted as an index of responsiveness

to the environment. High numbers of responses correlate with

intelligence and productivity, while a low number of responses is

suggestive of defensiveness, depression, and possibly malingering

(Exner, 1974).

The Rorschach variable X+% is the percentage of total

responses whose form conforms to the physical features of the

blot. It is considered one of the most important determinants and

is generally interpreted as a measure of the ability to perceive

the environment conventionally and realistically (Exner, 1974;

and Ogdon, 1977). An X+% below 70% is considered to raise

questions about perceptual accuracy and reality testing (Exner,

1974). Low percentages have been associated with the responses of

murderers (Wolfgang & Ferracuti, 1967) and proved useful in

discriminating property versus homicide offenders (Shagoury,



The Rorschach variable M refers to the presence of

human movement in a percept and has been a frequently researched

Rorschach variable (Exner, 1974). The frequency and quality of

M have been interpreted as representing internalization,

empathy, inhibition of impulses, and the ability to bridge inner

resources with external reality (Exner, 1974). It is considered

an index of emotional development since children exhibit low

frequencies (Ames, Metraux, & Walker, 1971). Davids (1973)

reported that the frequency of M discriminated between behavioral

aggression ratings of institutionalized boys.

When M is associated with bad form (M-), the likelihood

of psychopathology increases (Beck, 1965; Phillips & Smith,

1953) and suggests deficient social skills and poor interpersonal

relationships (Weiner, 1966). The presence of M- with any

notable frequency (Phillips & Smith, 1953; Weiner, 1966) or

when the ratio M-/M is greater than 1:3 (Phillips & Smith,

1953) suggests an increasing probability of psychosis.

The percentage of human responses, h%, reflects interest and

sensitivity to others (Exner, 1974). Rorschach M and h% were found

to aid in discrimination between a homicide group and a property

crime group (Shagoury, 1971). The ratio W/M, an index of aspirations

versus capacity (Exner, 1974) and was found to be the most powerful

discriminator between aggressive and nonaggressive alcoholics (Haramis

& Wagner, 1980).

The use of color as a response determinant is often

interpreted as an index of affective control and as related to


expression of anger or aggression (Klopfer, Ainsworth, Klopfer,

& Holt, 1954; Phillips & Smith, 1953; Schaefer, 1948; Exner,

1974). The use of color was one of the indices which

discriminated between behavioral aggression ratings of

institutionalized boys (Davids, 1973).

Rorschach weighted Sum C, which is coded as the sum

(1/2FC + CF + 1.5C), represents the sum and quality of the

subjects color responses. Sum C is generally interpreted as

reflective of responsiveness to emotional features of the

environment (Ogdon, 1977). Sum C has been found to be a

useful variable in discriminating between groups of defendants

some of which received psychological evaluation or raised the

insanity plea (Boehnert, 1983) Sum C also was useful in

discriminating between a property crime and a homicide group

(Shagoury, 1971).

The actual content of what is perceived has also been subject

of research with respect to aggressive behavior. A wide variety of

measures of pathological or hostile content have been used in the

literature (i.e., Elizur, 1949; Kane, 1955; Wolf, 1957; Sommer &

Sommer, 1958; Goldfried, Stricker, & Weiner, 1971; Rose &

Bitter, 1980; Shagoury, 1971). Although little or no research

is available contrasting these indices, certain contents have

been hypothesized to correlate with aggressive or deviant


Percepts with a content of blood, sex, religion, food, and

anatomy percepts are believed to be related to sexual,


aggressive, and primitive needs and impulses (Exner, 1974;

Phillips & Smith 1953; Rapaport et al., 1946). The content of

blood is believed to reflect sadistic and destructive impulses

(Phillips & Smith 1953; Rapaport et al., 1946), and sensitivity

and concern with the expression of destructive impulses (Phillips

& Smith, 1953; Rapaport et al., 1946; Davids, 1973). Religious

content is infrequent and believed to be associated with

preoccupation with good and evil and represent displacement of

sexual preoccupation and guilt (Phillips & Smith, 1953). Sexual

content is also rare in nonpatient samples and reflects

psychopathology. Blood responses are frequent in the records of

schizophrenics (Phillips & Smith, 1953).

The Rorschach has been successful at discriminating between

groups of aggressive and nonaggressive individuals. Shagoury

(1971) used a combination of psychological variables in a

discriminant function analysis to differentiate between a

homicide group and a property crime group. Rorschach Sum C,

pathological content, M, negative form level, and the percentage

of human responses were reported to be the most sensitive


Rorschach protocals of alcoholics were analyzed using a

stepwise discriminant function analysis (Haramis & Wagner,

1980). The procedure resulted in the accurate classification of

83% to 87% of the alcoholics into aggressive and nonaggressive

categories. The most powerful discriminator was the ratio W/M.


In another study (Rose & Bitter, 1980), a destructive content

scale was reported to help discriminate between groups of

released offenders which did well in the community for three

years and rapists who reoffended within six months of their


Minnesota Multiphasic Personality Inventory (MMPI)

The MMPI is a 566 item true-false personality

instrument developed at the University of Minnesota in 1941 by

Hathaway and McKinley (Dahlstrom, Welsh, & Dahlstrom, 1972).

It is probably the most frequently administered and researched

personality assessment instrument (Greene, 1980). It is used

cross-culturally and cross-nationally (Butcher & Pancheri,

1976). Ease and economy of administration have made the MMPI a

frequently used assessment instrument in prison settings

(Megargee & Bohn, 1979).

The MMPI yields a profile which consists of three validity and

ten clinical scales. The three validity scales, which assess

test-taking attitude, consist of the "L" (Lie), "F" (Frequency or

Infrequency), and "K" (Correction) scales.

The ten clinical scales were derived in an empirical manner to

discriminate between a criterion group of individuals meeting

certain diagnostic criteria and a control group taken from the

Minnesota population (Greene, 1980). The ten scales consist of

"Hs" (Hypochondriasis--Scale 1), "D" (Depression--Scale 2),

"Hy" (Hysteria--Scale 3), "Pd" (Psychopathic deviate--Scale 4),


"Mf" (Masculinity-femininity--Scale 5), "Pa" (Paranoia--Scale

6), "Pt" (Psychasthenia--Scale 7), "Sc" (Schizophrenia--Scale

8), "Ma" (Hypomania--Scale 9), and "Si" (Social Introversion--

Scale 0).

Research utilizing the MMPI has been voluminous. Its frequent

use with correctional populations for screening and

classification has stimulated much research using the instrument

to discriminate between groups of offenders (Megargee & Bohn,


An attempt to discriminate between inmates who had committed

violent offenses and those who had not was made by applying

discriminant function analysis to 141 MMPI profiles of adult male

inmates at a maximum security prison (Jones, Beidleman, &

Fowler, 1981). An equation based on elevations on MMPI scales F,

6, 7, and 8 correctly classified 72.9% of the violent inmates and

80.6% of the nonviolent inmates. Significant differences between

violent and nonviolent groups on these scales have been reported

in a number of studies (e.g., Oliver & Mosher, 1968; Panton, 1959,

1962; Potash, 1956).

Correct classification of 95% of 80 male prisoners into a

group with an arrest record of two or more assaults and a group

which did not have such a record was achieved by the use of a

neuropsychological test battery (Spellacy, 1978). Use of the

MMPI alone to discriminate between these two groups resulted

in a correct classification rate of 79%. The inmates

with previous records of assaults were characterized by


higher elevations on scales F and 6, and lower scores on scales K

and 5.

Scale score differences between a group of inmates with histories

of violent arrests and those with nonviolent arrests were also

reported by Deiker (1974). Profiles of 168 male prisoners in a

county jail or state correctional institution were analyzed and

yielded significant differences between a group of inmates with

crimes of threat, murder, or battery and a group with mostly

property offenses on scales F, K, 4, 7, 8, and 9. Deiker

suggested that a naysaying response bias may have contributed to

these results.

The MMPI profiles of 450 male and female offenders found guilty

and referred for a dispositional evaluation were examined by

McReary (1976). Assaultive offenders were found to have

significantly higher scale 9 scores than nonassaultive offenders.

A larger percentage of offenders with a 4-8/8-4 profile type were

assaultive than offenders with other profile types (4-3, 4-2, and


Lothstein and Jones (1978) found that elevations on scales

F, 4, 6, 7, 8, and 9 characterized a group of adolescent prisoners.

The profile type 8-4 was characteristic of the violent

adolescents. In another study, the sum of scales F, 4, and 9 was

1.5 standard deviations greater in a sample of juvenile

delinquents than a sample from the general population. This sum

demonstrated a reliability coefficient of 0.78 in a study of 426


nineteen year olds in the general population, indicating the

profile was stable across time (Huesman, 1978).

Intelligence Test Results

The hypothesis that intelligence interacts with personality

variables was advanced by Heilbrun (1979) who suggested that

psychopathy predicts violence for less intelligent criminals. Heilbrun

found that in a sample of 76 white Georgia state prisoners 76% of the

offenders with a low IPAT IQ and high psychopathy scores had been

charged with murder or rape. Holland (1981) reported a failure to

replicate Heilbrun's findings. Hinton (1983) reviewed the literature

concerning the predictive validity of intelligence with respect to

violent behavior and found conflicting evidence.

Summary of Test Correlates of Violent Behavior

Classification rates using psychological tests have yielded more

promising results than the limits of prediction suggested by the

early studies reviewed by Monahan (1981).

The studies reviewed suggest psychological test indicators such

as Rorschach color responses, human movement and content, form

quality, and pathological content correlate with aggressive

behavior. Studies utilizing the MMPI as a predictor consistently

report elevations on scales F, 4, 6, 7, 8, and 9 and low scores on

scale K. The use of intelligence measures as a predictor has

yielded less consistent results.


Cluster Analytic Approaches to Classification

One approach to identifying subgroups of offenders has been

the use of cluster analysis of MMPI data. This approach was

successful at identifying subgroups of individuals differing with

respect to type of crime committed and differing with respect to

institutional behavior including violence. Megargee and Bohn's

cluster analytically derived classification system (1979)

stimulated considerable validational research, which suggested

that this method of classification compares favorably with other

offender classification systems. Cluster analysis of MMPI

profiles has also been successful at identifying subgroups of

murderers (Anderson & Holcomb, 1983).

Megargee and Bohn (1979) developed a classification system

of criminal offenders by performing cluster analysis of offender

MMPIs. The classification system consisted of 10 groups based

on MMPI profile characteristics. The groups were derived from a

hierarchical profile analysis of three groups of 100 profiles

each from inmates at a federal medium security facility in

Tallahassee, Florida. The classification system was refined and

cross validated using 1214 additional inmates. Significant

differences between the groups for the number of subjects in

each group who were involved in violent and nonviolent

disciplinary infractions were found. Significant differences

between the groups were also found for reincarceration rates.

Edinger (1979) replicated these findings using 2000 male federal

prisoners and 1500 female state prisoners. Several later studies


used cluster analysis to classify individuals in other settings,

replicating some of Megargee's original groups and identifying

new ones (e.g., Quinsey, 1980; Nichols, 1979; Mrad, Kabacoff,

& Duckro, 1983).

Some failures to replicate between group differences of

Megargee's groups with respect to institutional behavior have

been reported. Megargee's classification system was not effective

at predicting which of 520 inmates from a federal penitentiary

would exhibit antisocial or aggressive behavior (Louscher,

Hosford, & Moss, 1983). Although significant between group

differences were found in number of disciplinary reports, no pair

wise differences emerged.

The relative efficacy of 4 classification systems, including

Megargee's, in predicting inmate institutional adjustment in a

penitentiary setting was examined by Hanson, Moss, Hosford, and Johnson

(1983). Hanson et al. examined demographic variables,

Megargee typology, security designation, and custody

classification data on 337 male inmates. Canoconical

correlations indicated the single best measure of overall

institutional adjustment was total number of disciplinary

reports. The best predictor of disciplinary reports was custody

classification. Being the member of specific Megargee groups was

the single best predictor for days in disciplinary segregation,

good time forfeited, and for positive work ratings.


Studies of Predictors of Institutional Aggression

Monahan (1984) indicated that studies of aggressive behavior in

acute care psychiatric facilities offer valuable data in which to judge

the short term predictive accuracy of psychological and psychiatric

prediction. Studies of aggressive behavior in institutional

settings have provided useful information on aggression within

controlled environments. These studies have provided information

on environmental variables, patient characteristics, and

behavioral correlates of violent behavior.

Studies of institutional aggression have indicated that

environmental variables such as the amount of structure and

activity influence the frequency of aggressive behavior. Age,

race, diagnosis, type of admission, and behavioral ratings of

agitation and psychotic behavior were found to be correlates of

institutional aggression.

A series of four studies which included 720 male inmates, 16

institutional subsettings, 63 types of infractions, and 30 types

of institutional sanctions was conducted by Edinger and Auerbach

(1978). A factor analysis of situations yielded a differentiation

between situations which had a high amount of staff supervision,

were structured, and in which staff-inmate interaction was

controlled and situations which were characterized as free-time

subsettings. A higher probability of assaults in free settings

than supervised settings was reported.

Although Rogers, Ciula, and Cavanaugh (1980) found a low overall

incidence of aggressive behavior in a 42 bed maximum security


psychiatric unit, they found that time of day and availability of

professional staff were correlated with the frequency of

aggressive incidents. Peak times for aggressive and socially

disruptive incidents were shift changes, meals, and periods of

concentrated treatment programming.

Mealtimes and the beginning of the day were also peak periods

for incidents in a study of three psychiatric hospitals which did

not admit "persistently violent individuals" (Fottrell, 1980).

A few patients were reported to account for the majority of the

incidents. Patients who acted violently towards themselves or

others were younger and more often diagnosed schizophrenic. A

study of battery incidents in a maximum security state hospital

also found that most batteries occurred in the daytime on the way

to meals (Deitz & Rada, 1982).

In the Deitz and Rada study (1982) batterers were reported to

have a longer mean length of hospitalization, were more

frequently prison transfers, and were more frequently nonwhite.

Race was also found to be a significant variable in a study of

5000 incidents occurring over the course of a year at a large

state hospital. Evenson, Sletten, Altman, and Brown (1974) found

that non-whites had higher risk rates for assault and antisocial

behavior. They found that young, unmarried males with deferred

diagnoses had the highest probability of engaging in assaultive


Some studies have examined short term behavioral correlates of

violent behavior within institutions. Few (7%) of the 5164


patients in two state hospitals studied by Tardiff and Sweillam

(1982) assaulted anyone in a three month period. Behavioral

ratings of these patients indicated assaultive patients

were more severely impaired on ratings of such psychotic symptoms

such as delusions, hallucinations, inappropriate affect, bizzare

habits, rituals, and exhibited more antisocial behavior than

nonassaultive patients. Assaultive patients were younger

and more frequently diagnosed nonparanoid schizophrenic, organic brain

syndrome, mentally retarded, or personality disordered.

Behavioral ratings and violent behavior was also examined by

Yesavage, Werner, Becker, and Mills (1982) in a comparison of

voluntary versus involuntary admissions to a 20 bed V.A.

inpatient psychiatric unit during the first week after admission.

Patients who were assaultive scored higher on behavioral

ratings of anxiety, conceptual disorganization, tension,

mannerisms, grandiosity, hostility, suspiciousness, motor

retardation, unusual thought content, and excitement than did

nonassaultive patients.

Hostile verbal behavior was used by Werner, Yesavage,

Becker, Brunsting, and Isaacs (1983) in an effort to predict

assaultive behavior by 110 schizophrenic V.A. inpatients. The

authors found that 32% of patients who engaged in hostile verbal

behavior committed an assault.

Type of admission has also been associated with assaultive risk.

In the Yesavage et al. (1982) study involuntary admissions were


rated more hostile than voluntary admissions. Sixty-five percent

of the involuntary admissions versus 47% of the voluntary

admissions incurred a violent incident in the first week of

admission. Rofman, Askinazi, and Fant (1980) compared the

records of 59 involuntary admissions to a V.A. inpatient unit to

59 voluntary admissions during the first ten days of admission.

Forty-one percent of the involuntary admissions versus 8%

of the voluntary admissions were involved in assaultive


Prediction Issues

The ability to predict behavior or events is a fundamental test

of knowledge. De Groot, quoted in Pedhauzer (1982), stated "If

one knows something to be true, he is in a position to predict,

where prediction is impossible, there is no knowledge" (p. 40).

Although prediction is not identical with explanation, and

may be far removed from causality, it is unlikely meteorologists

would make the evening news solely explaining yesterdays weather.

The importance of predicting violent behavior cannot be

underestimated since such predictions involve the civil liberties

and freedom of individuals. Predicting which individuals would be

violent under what circumstances has been called the paramount

consideration in the interface between mental health and the law

by the president of the American Psychiatric Association (Stone,



The accuracy of such predictions presents important ethical

and legal issues. Factors affecting accuracy involve the

determination of cutting scores and decision rules. While it may

be acceptable to adopt a criterion of more likely than not (51%)

for short-term civil commitment proceedings (Mental Health Law

Project, 1977), the application of a criterion of "a greater than

average probability" (greater than the overall base rate) would

lead to the deprivation of liberty of a large number of people

(Monahan, 1981).

The selection criterion employed to determine cut scores

determines the rate of true positives to true negatives and the

absolute number of successful predictions. Monahan (1981) urged

that regardless of the selection procedure used, "by all means

the rule should be made explicit" (p. 38).

Task forces of the American Psychiatric Association (1977) and

the American Psychological Association (1978) have maintained

that psychiatrists and psychologists do not possess adequate

information or scientific knowledge to predict violent behavior.

Monahan (1981) suggested the question is not whether or not

predictions of violence could be made, nor whether or not they

should be made, but the question should be how accurately can

they be made and in what circumstances.

The method by which predictions are made is the subject of the

debate between clinical and statistical prediction (Meehl, 1954).

Meehl reviewed the research data in a variety of fields and found
"overwhelming" evidence in favor of statistical prediction. He


found little need to retract "95%" of his position 30 years

later (Meehl, 1986).

Kastermeir and Eglit (1973) suggested that the resistance to

the idea of statistical prediction in the area of prediction of

violent behavior stemmed from the view that legal issues are

intrinsically individualized, the fact that actuarial methods

explicitly acknowledge that errors will be made, uneasiness over

stating the reasons decisions which are made which run counter to

statistical predictions, and concern about loss of status and

jobs to clerks armed with statistical formulae.

To this list Monahan (1981) added that concerns about stating

certain reasons for decisions which are statistically made (i.e.,

race and sex) lead to resistance. Einhorn (1986) indicated that

the unavailability of the required data to make a statistical

prediction, or insufficient time (in the case of emergency

commitment procedures), results in a necessity to rely on clinical

experience and judgement. Increased overall accuracy, explicit,

empirically testable rules, and increased consistency in decision

making are cited as advantages of statistical prediction

(Einhorn, 1986).

Monahan (1981) suggested the issue is not one of clinical versus

statistical prediction, but a question of what can the clinician

do to increase the accuracy of prediction. To accomplish this end

Monahan suggested that determination of the base rate of violence

in the population to which the predictions are to be applied

needed to be made the primary consideration. Megargee (1976)


maintained a similar position and suggested that mental health

professionals should limit themselves to predicting violent

behavior to populations with high base rates of violent behavior

such as those who have already exhibited violence.

Monahan (1981) added that obtaining information on a limited

number of reliable valid predictive relationships was necessary

to limit the reliance on illusory correlations. Monahan (1984)

urged a shift away from clinical attempts at long term

predictions of dangerousness in chronically institutionalized

populations and a shift toward a multidimensional approach to

examining short term predictions.

Statement of the Problem

The early research on attempts to predict violent behavior

focused on long term prediction using chronically

institutionalized samples (Monahan, 1984). Monahan (1984)

suggested that this research has been overgeneralized and that

empirical approaches using a multivariate model which attempt to

make more short term predictions in less chronic populations may

be more fruitful.

Many predictive efforts have failed due to low base rates

(i.e., Michigan Department of Corrections cited in Monahan, 1981;

Steadman & Cocozza, 1974; Wenk, Robison, & Smith, 1972).

Predictive efforts are most likely to be successful when the

target behavior has an approximate 50% base rate (Meehl, 1954).

Previous studies indicate involuntary admissions (Yesevage


et al., 1982; Rofman et al., 1980) and those adjudicated

incompetent to stand trial (Steadman & Cocozza, 1976) have

relatively high base rates of aggressive behavior.

The present study attempts to use a multivariate model which

includes demographic data, arrest history, diagnosis,

institutional variables, and test data to develop multiple

regression equations which identify groups of individuals with

higher than base rate involvement in institutional incidents and


The need for efficient classification systems in forensic

facilities to aid in management and treatment remains (Mrad,

Kabacoff, & Duckro, 1983). Classification systems based on

psychological test results have been as effective or more

effective than clinical predictions at discriminating between

potentially aggressive and nonaggressive groups (Shagoury, 1971;

Spellacy, 1978; Haramis & Wagner, 1980; Jones et al., 1981).

Cluster analysis of MMPI profiles has been successfully

employed in Megargee's MMPI based system for offender

classification (Megargee & Bohn, 1979) and in developing an

MMPI typology of murderers (Anderson & Holcomb, 1983).

Megargee's method has been successfully used to identify

naturally occurring groups of adult offenders (Nichols, 1979), and

half-way house residents (Mrad et al., 1983). Implementation of

Megargee's system was successful at reducing institutional

violence in a medium security federal prison (Megargee & Bohn,



Cluster analysis of the MMPI profiles of forensic patients may

identify new groups which are likely to be unique to forensic

psychiatric facilities. The present study attempted to identify

naturally occurring MMPI subgroups in this population and to

validate the groups by demonstrating between group differences in

demographics, institutional behavior, and independent

psychological test data.

Individuals found incompetent to stand trial have been

relatively ignored in forensic psychological research (Steadman,

1984). A secondary purpose of the present study was to provide

more data on a population of forensic inpatients consisting

predominantly of individuals adjudicated incompetent to stand

trial. Multiple regression procedures were used in an

exploratory investigation of the predictors of the length of time

required to return to competency. It was hoped this would add to

the sparse literature on the subject (e.g., Cuneo, Brelje,

Randolph, & Taliana, 1982; Heller, Traylor, Ehrlich, & Lester,


The hypotheses tested in the current study were as follows.

1. Some weighted combination of psychological test indices and

demographic data derived from multiple regression analysis could

identify residents at higher than base rate risk for becoming involved

in institutional incidents.

2. A similarly derived equation could identify residents at risk

for a) aggressive incidents, b) fights, and c) incidents requiring use

of force.


3. Some combination of psychological test indices and demographic

data could predict length of treatment for those residents

adjudicated incompetent to stand trial.

4. Naturally occurring groups of forensic inpatients could be

identified by hierarchical cluster analysis of residents MMPI data.

5. The derived groups would be valid in that significant

differences would be found between the groups on demographic

data, institutional adjustment, and other psychological test




The present research used demographic and psychological test

variables to predict institutional incidents, aggression, and

length of stay for residents in a state forensic psychiatric

facility. The present study also tested the hypothesis that

naturally occurring groups of forensic patients could be

identified by cluster analysis and that the groups would be

significantly different on demographic, test, and behavioral

characteristics. The subjects, institution, and variables used

are discussed below.


Subjects were 451 male residents from a pool of 474 residents

evaluated by the center's Psychology Service between March, 1981, and

April, 1985. Incomplete medical records, conflicting identifying

information, and lost psychological or medical records resulted in the

exclusion of 23 of 474 (4.85%) residents from this study. Residents'

ages ranged from 15 to 64, with a mean age of 30.52.

The Psychology Service was staffed by two full time licensed

Ph.D. clinical psychologists and two assistants who conducted

psychological evaluations as part of a contract with the


Department of Clinical Psychology at the University of Florida.

The assistants were pre-doctoral graduate students with similar

training in psychological test administration and scoring in that


The Institution

North Florida Evaluation and Treatment Center is a 200 bed maximum

security treatment facility for mentally disordered offenders. Ten

buildings house from 18 to 27 residents. The center is comprised of

four units one of which houses residents participating in a sex

offender treatment program. These sex offenders were not included in

the present study. The 451 residents in the present study consisted of

residents on the three units comprised mostly of residents who had been

adjudicated incompetent to stand trial. A small percentage of those

found not guilty by reason of insanity and transfers from the

department of corrections were also residents on these three units.

Evaluations of residents were routinely completed by the

Psychology Service within the first 30 days of admission.

The Psychology Service files contained evaluations of 474

residents. The evaluation consisted of a clinical interview and

testing. Some tests which were routinely administered included

the MMPI, WAIS-R (Wechesler, 1981) or rarely the Peabody Picture

Vocabulary Test (Dunn, 1965), projective drawings, and the

Rorschach. Variability in the tests administered existed due to

reading difficulties, limited cooperation, and time constraints.


Materials and Measures

Demographic Variables

The selection of variables used in the present study was based on

previous research findings, and the availability of such data in

existent records. The selection of demographic data to be used in

prediction equations was determined by the previous prediction

literature. An effort was made to use nonredundant summary variables

to minimize the number of independent variables thereby increasing the

stability of the solutions (Pedhazur, 1982).

Admission status was coded as incompetent to stand trial,

not guilty by reason of insanity, prison transfer, and other.

Individuals adjudicated incompetent to stand trial were court

ordered to one of two Florida maximum security psychiatric

facilities if they were also adjudicated to be a danger to

themselves or others and in need of treatment in a secure

facility. Type of admission has been associated with between

group differences in institutional aggressive behavior (Deitz &

Rada, 1982; Yesavage et al., 1982).

Age was coded as the subject's age at time of testing. Race was

coded as white, black, Hispanic ethnic group, and other. Age and

race have been consistently reported to correlate with aggressive

behavior (i.e., Deitz & Rada, 1982; Sweillam, 1982).

Diagnosis was coded from resident's NFETC official discharge

DSM-III diagnosis (American Psychiatric Association, 1980).

Diagnosis has frequently emerged as associated with aggressive


behavior (Evenson et al., 1974; Fottrell, 1980; Tardiff &

Sweillam, 1982) although methods of diagnosis and findings have

not been entirely consistent.

After reviewing the frequencies of each type of diagnosis it was

decided that diagnoses would be coded as no axis-I diagnoses, affective

(for DSM-III major affective disorders, n = 19, or schizoaffective

disorder, n = 8), paranoid schizophrenia, all other types of

schizophrenia, and "other" (this consisted of schizophreniform

disorder, n = 5, post-traumatic stress disorder n = 1, dysthymic

disorder, n = 5, brief reactive psychosis, n = 14, and paranoid

disorder n = 6). Organic brain syndrome (OBS) was coded if the

discharge diagnosis included dementia, epilepsy, or organic

personality syndrome.

The NFETC discharge diagnoses included few diagnoses of

personality disorder and other diagnoses except mixed personality

disorder (n = 39) and antisocial personality disorder (n = 51).

It was decided personality disorder would be coded as presence of

any Axis-II diagnosis exclusive of antisocial personality disorder

(n = 79). Presence of antisocial personality disorder was coded


The presence of substance abuse or dependence was also coded

separately and was coded as positive if the individual had a

diagnosis of substance abuse or dependence, a history of two or

more alcohol or drug related arrests, or a previous history of

treatment for alcohol or drug dependency.


Marital status and education were coded from medical records.

It was judged that a reliable source of residents employment

status was not readily available. Employment status was not


Arrest Data

National Crime Information Center (FBI-NCIC) and Florida

Department of Law Enforcement (FDLE) arrest records were routinely

obtained upon an individual's admission to NFETC. These records were

the source for determining total number of arrests and number of

violent arrests. The age of first arrest was obtained from arrest

records, medical records and reports. Total arrests, violent arrests,

and first recorded arrest were coded by two assistants if they occurred

in Megargee's list of violent offenses (Megargee, 1982).

As a test of the reliability of the assistant's coding the

primary investigator recorded random samples of each assistants

arrest codings. Inter-scorer reliability was assessed between the

principal investigator and Assistant One with r = .9742 (n = 22)

for total arrests, and r = .98333 (n = 22) for violent arrests.

For records coded by Assistant Two and recorded by the principal

investigator, r = .9309 (n = 23) for total arrests, and r = .9827

(n = 23) for violent arrests. These results indicated that the

codings were highly reliable and certainly sufficient for

purposes of the present study.

Test Data

Residents were routinely referred to Psychology Service for

evaluation within four weeks of admission. Medical records were

reviewed, and residents were interviewed and tested individually.

Residents were informed that a report would be included in their NFETC

records. Among tests routinely administered by psychology services

were the MMPI, WAIS-R, and the Rorschach.


Psychology services administered two forms of the MMPI to

residents. One hundred eighty-three subjects were administered Form-R

of the MMPI and 110 residents were administered the MMPI-168 (Overall,

Higgins, & DeSchweinitz, 1976), a short form of the MMPI. Due to

criticism concerning the equivalency of the two forms (Hoffman &

Butcher, 1975; Ward, Ward, & Moore, 1983) and lack of data concerning

the use of the MMPI-168 in institutionalized forensic populations

(Stevens & Reilley, 1980), 40 randomly chosen Form-R MMPI's were

recorded as MMPI-168s and the resultant K-corrected t-scores compared.

Correlations between Form-R scored and the MMPI-168 scored

profiles yielded correlation coefficients between .37739 and

.93273 (see Table 1). Significant differences were found between

the differently scored profile for K-corrected scores on scales

Hs, Hy, Pd, Mf, Pa, and Sc (see Table 1). On the basis of these

results, MMPI-168s were excluded from the present study to

reduce measurement error.

Table 1

Mean K-Corrected T-score Differences and Correlations between
MMPI Form-R recorded as MMPI 168: Paired Observations

Form R
Score *
Scale 168 Score S. D. t prob. correlation


































































value (one-tailed, p = .05) = .26406



An additional 16 MMPI profiles from residents admitted after April

1985 were included in the cluster analysis to increase the power of the

analysis. Data from residents of the additional 16 profiles were not

included in the multiple regression equations since the full set of

data was not collected. These additional 16 profiles were included in

calculation of the overall mean NFETC MMPI profile.

A subset of MMPI scales was used in the multiple regression

analyses. Scales F, K, 4, 6, 8, and 9 were selected for inclusion

since the literature reviewed suggested these scales were most

consistently found to differ, or discriminate between, aggressive

and nonaggressive groups. All MMPI scales were not used in order

to minimize the variable to sample size ratio, thereby increasing

the stability of the solutions (Pedhazur, 1982).

Intelligence Measures

Psychological Services administered the Wechesler Adult

Intelligence Scale-Revised to residents unless time constraints

precluded administration, the residents mental status at the time or

spoken English would not yield a valid estimate, or if a valid recent

intelligence estimate was available.

Intelligence quotient (IQ score) was coded as the best WAIS-R

estimate available from testing or medical records. The WAIS-R yields

a Full Scale I.Q. (FSIQ), a Verbal I.Q. (VIQ) and a Performance I.Q.

(PIQ). The VIQ is calculated from the subscales of Information,

Arithmetic, Vocabulary, Comprehension, and Similarities. The PIQ is


calculated from the subtests of Picture Completion, Picture

Arrangement, Block Design, Object Assembly, and Coding.

The WAIS-R subtest correlations with each other range from .33

(Digit Span and Object Assembly) to .81 (Vocabulary and Information).

Verbal and Performance IQ's correlate .74. VIQ correlates .95 with

FSIQ, and PIQ correlates .91 with FSIQ.

Since the full WAIS-R was not administered in some cases

prorating or estimation of scores was based on the subtests

administered. Coded IQ was, in order of preference, WAIS-R FSIQ

or prorated FSIQ, WAIS-R VIQ or pro-rated VIQ, WAIS-R PIQ or

prorated PIQ, and previously reported WAIS-R IQ.

Conflicting evidence for the role of intelligence in predicting

recidivism or aggressive behavior exists (Heilbrum, 1979; Hinton,

1983). Heller et al. (1981) found that intelligence test results

were predictive of length of treatment required to restore

competency to stand trial.


The Rorschach was administered and scored using Exner's

Comprehensive Rorschach System (Exner, 1974). The following variables

were selected for use in the multiple regression equations based upon

the previously reviewed literature. Except where noted, the variables
were coded according to Exner's (1974) scoring criterion.
Rorschach R was defined as the total number of responses to

the cards. The Rorschach variable X+% is the percentage of


total responses (excluding pure Color responses) whose form conforms to

the features of the blot and represents good form quality. The

determination of whether or not a percept conforms to a particular

region of the inkblot is primarily determined by the use of frequency

tables (Exner, 1974).

The Rorschach variable M refers to the presence of human

movement in a percept. The frequency and form quality of

M were coded. Rorschach M+ indicates the human movement response

conformed to the features of the blot, whereas M- indicates

the human movement response did not conform to the features of

the blot. To facilitate computation and summarize the quality of

the human movement response, the ratio M- + 100 / (M- + M+) + 100

was used in the present study.

The ratio of responses which include the whole blot, to the

total number of responses which include human movement, W/M

is generally interpreted as an index of aspirations to

current capacity (Exner, 1974). Rorschach W/M was coded as W + 100 /

(M+ + M-) + 100. The percentage of responses which include human

percepts as content was coded as h%.

The use of color as a response determinant is often examined

as an index of affective control and related to expression of

anger or aggression. Responses that include color as a

determinant are coded as FC, CF, or C depending on the relative

weight the individuals responses indicated that either the color

of the blots or the form was primary in determining the percept.

Pure C is coded when the percept is solely based on color as


determinant. The rare Color naming response (e.g., "red") was also

coded as C in the present study.

Rorschach Sum C, which was coded as the sum (1/2FC + CF + 1.5C),

represents the sum and quality of the subjects' color responses. An

index of M to Sum C (EB) was computed as (M+ + M-) + 100 / Sum C + 100.

The ratio EB is reflects the degree to which the person is more prone

to use inner life versus interaction with the world for satisfaction of

important needs (Exner, 1974).

For purposes of this study path% was defined as the ratio:

b + sex + rel + fd + an / R. Percepts with a content of

blood, sex, religion, food, and anatomy percepts are believed to

be related to sexual, aggressive, and primitive needs and

impulses (Exner, 1974; Phillips & Smith 1953; Rapaport, Gil,

& Schaeffer, 1946). This index was created by the present

author since it could be readily coded from Exner's content

categories. Previously published methods of scoring pathological

content were not used since this would require substantial

recoding, often from illegible transcripts of subjects initial

verbal responses.

The interscorer reliability of two of the examiners which

administered and scored many of the Rorschach protocals in the present

study has been reported in a previous study (Unger, 1985). Every fifth

response of 30 randomly chosen protocals from a pool of 150 student

protocals were scored by both examiners. Unger (1985) reported

interscorer reliability agreement of 92% for overall responses, 98% for

location, 94% for determinants, and 93% for form quality (X+%).


Dependent Measures

Dependent Variables

The primary dependent measures in the multiple regression

equations were coded from NFETC Incident (IR) and use of force (UFR)

reports (Appendix A). These reports were implemented by NFETC in

September 1982. Staff at the center are required to complete an

incident report form for each time a resident threatens or actually

harms another person or property. These reports are reviewed and

signed by supervisors and security for completeness and accuracy.

Time of incident, circumstances, and response to incidents are

coded on the incident report by checking off appropriate

categories. Incident reports are divided into five sections. The

first section indicates the particular people involved in the


Section two indicates the location of the incident.

Section three contained the majority of the data of interest in

the present study. The type of incident was coded on incident reports

as A. Refusal to take medications; B. Violation of standing

procedures; C. Refusal to comply with verbal orders; D. Verbal abuse

toward staff or other residents; E. Resident threatened violence

(toward self or other); and F. Resident performed violence (toward

self or other). Residents response to security was also coded as to

whether the resident complied, resisted verbally, or resisted

physically. More than one problem could be coded during any



Staff response to the incident was coded on incident reports as

A. Verbal orders issued to resident; B. Physical force applied; C.

Condition B watch; D. Placed in observation room (Condition A watch);

E. Nurse called; F. Security called; G. Placed in seclusion room; and

H. Restraints applied.

Building Incident Rates

A separate index file was maintained by NFETC security which

listed incident report numbers by resident. This included any incident

which listed the resident in section I of the report. This file was

used in the present study to calculate building incident rates in the

following manner.

Indexed incidents were totaled by building and month for

each resident at NFETC (exclusive of sex offender buildings)

between September 1982 and December 1985. For each resident in

the study a total of all incidents occurring in his building

during months he was present for more than 14 days was made. The

number of indexed incidents which involved the particular

resident was subtracted from this number and then divided by the

number of months of the residents stay yielding a building

monthly incident rate for each resident.

Individual Incident Rates

Each resident's total incident rate was calculated from raw

incident report and use of force report data which were collected,

stored, and supervised by NFETC security. In calculating each


resident's incident rate, reports which involved building searches,

reports in which the resident was not directly involved, minor medical

injuries, and reports in which the resident was clearly only the victim

of a verbal or physical assault were not included. Thus, this total

was below each resident's index card total. The total number of

incidents was divided by each resident's length of stay yielding a

monthly incident rate for each resident.

Violent incidents were coded as a subset of the total incident

figure above. An incident was coded as aggressive if it involved

violent physical contact with another resident, staff, or

security. The number of such incidents was divided by each

resident's length of stay to yield a monthly aggressive incident


On occasion, physical aggression would occur following the initial

incident intervention. These incidents were added to each resident's

total violent incidents to yield a monthly rate of extended aggressive

incidents. Thus, aggressive incidents formed a subset of extended

aggressive incidents.

Fights constituted a subset of violent incidents and were coded

from narrative reports. Reports of clearly identified victims of

attacks were excluded, and a monthly fight rate was calculated

for each resident.

Use of force reports were required whenever physical

intervention was required by staff or security to control the

resident and most likely constituted the most serious subset of


incidents. Use of force reports were tabulated by resident and

were used to calculate monthly use of force rate for each


Building Response

A measure of each building's response characteristics was derived

from examining the rate of physical intervention, restraint, or

seclusion in response to incidents which did not involve actual

physical violence by the resident.

One hundred ninety-four reports were identified which included

verbal aggression, but no physical aggression, or threats of

violence to self or property. These were tabulated by building

and the percentage of these reports which included physical

intervention, restraint, or seclusion by staff was calculated and

coded as building verbal aggression response.

Length of Stay

Resident's length of stay was coded as days between date of

admission and date of discharge. Length of stay was tabulated by

building to yield building average length of stay.


The selection of predictor variables was determined by previous

research findings and the availability of such information in

existing files. K-corrected Minnesota Multiphasic Personality Inventory

(MMPI) scales were used. The MMPIs were not included if the


psychological report indicated evidence of a random response style or

if the profile had a scale K T-score above 70 T. Eleven profiles were

excluded from cluster analysis and multiple regression analyses due to

K scores greater than 70 T.

Prediction equations were generated by sequential application

of stepwise multiple regression to blocks of predictor variables

(Cohen & Cohen, 1975). In this method, the stepwise regression

procedure selects variables from the subset of predictor

variables constituting the "block." At each stage the variable

which has the largest semipartial correlation is entered. When

no predictor variable from the block would make a further

significant contribution to the variance accounted for, the

analysis is terminated. If the addition of a variable results in a

previously entered variable no longer making a significant

contribution, as sometimes happens with highly correlated variables,

the previous variable is removed from the equation.

In a blockwise selection procedure surviving variables from

the previous block which had made a significant contribution are

added to the next block and the stepwise multiple regression

procedure described above repeated until all variables and blocks

are analyzed (Cohen & Cohen, 1975). Resultant equations are

obtained in an a posteriori order based solely on the relative

uniqueness of predictor variables in the sample at hand (Cohen

& Cohen, 1975).


Stepwise multiple regression procedures have been criticized as

being unsuitable for explanatory research. The use of a large number

of predictor variables relative to the sample size results in

capitalization on chance and make overall tests of multiple R-squared

invalid (Cohen & Cohen, 1975). These criticisms are less of a problem

when the research goal is entirely or at least primarily predictive

when the sample size is relatively large relative to the number of

predictor variables, and when the results are cross validated (Cohen

& Cohen, 1975).

A more realistic estimate of the multiple R-squared can be

calculated as a "shrunken R-squared," where shrunken R-squared =

1 ( 1 R-squared) X ( n 1 ) / (n k 1) (Cohen & Cohen,

1975), where n is the sample size and k is the number of

predictor variables. Two estimates of Shrunken R Squared may be

made. A liberal estimate involves calculating k as the number of

variables actually entered into the equation. A conservative

estimate is calculated defining k as the total number of

variables examined prior to the stepwise selection procedure

(Cohen & Cohen, 1975). An estimate of the magnitude of error in

estimating values in other samples is calculated by the standard

error of estimate which yields the estimated standard deviation

of residual errors.

Stepwise multiple regression was applied in blocks to generate

prediction equations for the criterion variables incident rate,

aggressive incident rate, extended aggressive incident rate, fight


rate, and use of force rate. All multiple regression analyses were

performed using Microstat software (Ecosoft, 1984).

The same procedure was used to generate a prediction equation for

length of stay for residents adjudicated incompetent to stand

trial. Prediction equations were then utilized to generate

classification tables for different incident types using various

cutting scores. The resultant classification rates were compared

with classification rates based upon chance (50%) and the overall

base rate.

The order of block entry was determined by availability of data,

difficulty and costs of obtaining the data, and finally by the

completeness of the data for all subjects. Since all subjects

did not complete all of the psychological tests, test data were

entered last.

Possible differences between groups of residents who were and

were not administered a Form-R MMPI and between those residents

who were and were not administered a Rorschach were examined using

t-tests. Differences between the groups of residents would pose

limits to the generalization of equations which included test


Categorical variables such as race, diagnosis, building, and,

admission type were coded as dummy variables (Cohen and Cohen,

1975) in the multiple regression analyses (See Table 2). All

analyses used the same order of block entry (Table 3).

Block 1 consisted of the basic demographic variables of age,

education, race, marital status, and admission type. Block 2

Table 2

Variable Codina for MultiDle Reqression Analyses

black, Hispanic (0,0) = white

Marital Status:

Admision Status:

Axis I:



married, divorced, other (0,0,0) = never married

not guilty by reason of insanity, other,
(0,0) = incompetent to stand trial

affective disorder, nonparanoid schizophrenia,
paranoid schizophrenia, other axisi,
(0,0,0,0) = no axis I diagnosis

Unit 2, Unit 3, (0,0) = Unit 1

blg. 6, b1g. 7, b1g. 8, b1g. 9, b1g. 10,
blg 13, b1g. 14, (0,0,0,0,0,0,0) = other


Table 3

Multiple Regression Blocks



age Axis I: b1g. b1g. total Rorsch: MMPI: IQ
avg. arrests
ed. affect. l.o.s. 6 R F
Race: o. schiz. b1g. 7 arrest X+ % K
black par. scz. rate 8 violent path % 4
Hisp. o. Ax I. b1g. 9 W/M 6
verbal current
Marital: personal. incid. 10 violent Sum C 8
disorder resp. arrest
married 13 M-/M 9
antisoc. Unit
div. II. 14 M/Sum C
other Unit H %
drug/etoh III.




consisted of discharge diagnoses. Although these diagnoses were

not predictor variables in the chronological sense, it was

assumed these diagnoses were descriptive of the residents and

representative of the syndromes and symptoms for which they were

admitted and treated.

Blocks 3 and 4 represented institutional variables. Block 3

consisted of each resident's building average length of stay,

building incident rate, building response to verbal aggression,

and the unit to which the resident was assigned. These data were

entered before Block 4, which represented the building to which

the resident was assigned, because Block 3 variables contained

more specific and meaningful data.

Block 5 data consisted of the arrest variables of total arrests,

violent arrests, age of first arrest, and presence of violent

current offense. These data were entered after the previous blocks

since it was likely that this type of data would be difficult to

obtain in many instances.

Psychological test data were entered last since all

residents did not complete all tests, the economic cost of

collecting such data, and to assess whether or not psychological

test data would contribute significant additional information

beyond demographic, diagnostic, and institutional data.

The Rorschach variables R, X+%, path%, M- to M, W to M, Sum C, M

to Sum C, and h percent were entered as block 6. Rorschach data were


entered prior to other test data since subjects who completed the

Rorschach represented a larger subsample.

The MMPI scales F, K, 4, 6, 8, and 9 were entered as block 7.

The MMPI data were included prior to the single variable block 8 (I.Q.

score) due to the relative costs of administering an MMPI and an

individually administered Wechsler Adult Intelligence Scale-Revised


Cluster Analysis

The hypothesis that naturally occurring groups of forensic

inpatients could be identified by hierarchical cluster analysis of

K-corrected MMPI T-scores was tested using 188 subjects administered

the test. The MMPI data were cluster analyzed using hierarchical

profile.analysis using Ward's (1963) method and the CLUSTAN (Wishart,

1978) computer program at the Northeast Regional Data Center.

The procedure is an agglomerative hierarchical method which uses

Euclidean distance as the similarity measure and minimizes within

cluster variance (Blashfield & Morey, 1980). Since cases are

not assigned to the nearest cluster after initial assignment by

this procedure, procedure RELOCATE, which does assign cases to

the nearest cluster, was performed.

If the resultant cluster groups are valid, significant

between group differences are expected (Blashfield, Aldenderfer, &

Morey, 1979). Following the RELOCATE procedure multivariate and

univariate anlayses of variance were conducted to test for


between group differences on continuous variables. Chi-square

analyses were conducted for categorical variables.

In a deviation from Megargee's cluster analytic procedure

(Megargee & Bohn, 1979) records with an F scale score greater

than 100-T were not eliminated from the present sample. Random

selection of 32 profiles from the present sample found nine

(35.5%) profiles with F scale scores greater than 100-T. These

profiles were included since they constituted a large part of the

sample to be studied. A large percentage (24%) of a halfway

house sample was eliminated by this procedure in a previous study

(Mrad et al., 1983).



Sample Characteristics

Demographic and Diagnostic Characteristics

The mean age of the 451 NFETC residents in the present sample was

30.5. The racial composition was mixed (Table 4), 44.1% of the

residents were white, 38.4% black, 16.6% were of Hispanic ethnic

origin, and 0.9% "other." The Hispanic group was comprised of 70.6%

residents of Cuban origin, 11% residents of South American origin, 11%

of Puerto Rican origin and 7% other Hispanic origins. Of the Cubans,

56.6% were refugees from Cuban prisons and asylums who arrived in the

United States during the 1980 Mariel boat lift. These refugees

comprised 40% of the entire Hispanic ethnic group.

Few residents were presently married (5.6%), and most had never

been married (69.2%). The mean number of years of education

(9.9) suggests that most of the sample had not graduated high

school. Demographic and diagnostic data descriptive of the

present sample are presented in Tables 4-8.

The majority of residents in the present sample had been

adjudicated incompetent to stand trial (91.4%). The overall

sample had a substantial criminal history (Table 5). Residents

had been previously arrested a mean 7.9 times with a mean

Table 4

Demographic Data

Variable Mean Sample S. D. Minimum Maximum N



















Marital Status:

unmarried 312

married 25

divorced 108

other 6










Table 5

Admission Type and Arrest Data

Variable Frequency Percentage

Admission Type:

Incompetent to stand trial

Not guilty by reason
of insanity

Baker Act transfer
from D. 0. C.


Type of Current Arrest:









Arrest History:

Mean Sample
S. D.

Minimum Maximum N

Total Arrests

Violent Arrests

Age at First Arrest








1.9 arrests for violent crimes. Although juvenile records

were rarely available, the mean age of first known arrest based

primarily on adult records was 21.8. The majority of residents

were faced with a violent current charge (67.4%). Approximately

10% of the sample had present charges of some form of homicide, and

14% of the sample had a present or past arrest for homicide.

The majority of the residents in the present sample were

suffering from severe forms of mental illness. The modal Axis-I

of diagnosis was schizophrenia (Table 6). Sixty-four percent of

the sample received discharge diagnoses of some type of

schizophrenia. Paranoid schizophrenics made up 41.24% of the

sample, and other forms of schizophrenia accounted for an

additional 22.84%. A substantial minority of residents (23.06%)

received no Axis-I discharge diagnosis.

Personality disorders other than antisocial personality disorder

were not frequently diagnosed (17.5%, Table 7). Antisocial

personality disorder was also not frequently diagnosed (11.3%).

The NFETC residents had frequent histories of alcohol and substance

abuse. Residents with either a discharge diagnosis which included

some form of substance abuse or dependence, history of two or

more substance related arrests, or history of treatment for

substance dependency included 27.7% of the sample. Discharge

diagnoses of organic brain syndromes, or organic personality

syndromes were rare (5.5%).

Residents were fairly evenly assigned to Units (Table 8).

Residents were assigned to Unit and building at admission.

Table 6

NFETC DSM-III Axis-I Discharge Diagnoses

DSM-III Diagnosis Frequency Percentage

0. No Axis-I Diagnosis (Below) 104 23.06

1. Major Affective Disorder 19 4.21

2. Schizoaffective Disorder 8 1.77

3. Paranoid Schizophrenia 186 41.24

4. Schizophrenia (Other) 103 22.84

5. Schizophreniform 5 1.11

6. Post Traumatic Stress Disorder 1 0.22

7. Dysthymic Disorder 5 1.11

8. Brief Reactive Psychosis 14 3.10

9. Paranoid Disorder 6 1.37

Note: For purposes of multiple regression analyses Major
Affective Disorder and Schizoaffective Disorder were combined
into "Affective Disorders", and diagnoses 5-9 above were
combined into the category "other Axis-I."

Table 7

Other DSM-III Discharge Diagnoses

DSM-III Diagnosis Frequency Percentage

Personality Disorders
(not including antisocial personality disorder)

not diagnosed 372 82.48

diagnosed 79 17.52

Antisocial Personality Disorder

not diagnosed 400 88.69

diagnosed 51 11.31

Organic Brain Syndromes

not diagnosed 426 94.46

diagnosed 25 5.54

History of Substance or Alcohol Abuse or Dependence

absent 326 72.28

present 125 27.72

Table 8

Unit and Building Assignments

Assignment Frequency Percentage




























Assignment was made on the basis of available bed space. Unit One,

Unit Two, and Unit Three housed 33.5%, 27.7%, and 38.8% of the sample

of residents respectively. Assignment to buildings was also fairly

evenly distributed, with the following exceptions. Residents from

Building 12 and 15 combined comprised only approximately 5% of the

sample. This was due to changes in the use of these buildings during

the time period of the present study. A somewhat higher proportion of

residents (20%) were assigned to Building 10.

Test Data

Test data indicated that NFETC residents were poorly functioning

cognitively and emotionally. The mean level of intellectual

functioning was at the very low end of the low average range of

measured intelligence (80.5). This was just slightly above the range of

IQ scores which are indicative of borderline intelligence (70-79).

The MMPI data were characteristic of acutely hospitalized

inpatient populations. The mean MMPI profile (see Table 9) is

presented in Figure 1. This profile included data from an additional

16 residents which were used in the cluster analysis, and eliminated

residents with a scale K score greater than 70 T since these were

considered invalid.

Rorschach data are presented in Table 10 along with means and

standard deviations of a nonpatient and a schizophrenic sample

(Exner, 1977). The low percentage of percepts conforming to the

physical features of the blots (59%) is indicative of poor

reality testing and is characteristic of psychotic individuals.

Table 9

NFETC MMPI Descriptive Statistics

K corrected

Scale Mean Deviation






























70 -- ------------------- ------ ------- ---------- -------

50 -------------------------------------------------------------

L F K Hs D Hy Pd Mf Pa Pt Sc Ma Si
n = 188

Figure 1. Mean Group NFETC MMFI Profile

Table 10

Rorschach NFETC Descriptive Statistics


n = 325

n = 210










6.58 3.7

* Exner (1974)


n = 237



X+ %

path %

Sum C

h %

M -

M +













































Incident Data

Security maintained an index file which listed incident report

numbers whenever a resident was mentioned in an incident report form.

This sometimes included reports in which the resident was only a

witness, only marginally involved, or part of a building wide search.

Indexed incidents were used in the present study to estimate monthly

incident rates on each building. Indexed incident data included all

residents assigned to the three units which did not treat sex


A total of 4466 index file incidents were tallied for the

period between September 1982 and December 1985. A single

incident could be tallied more than once, depending on the number

of residents involved, therefore the present results are an

overestimate of actual building incident rates.

Residents in the present study were those residents who

were tested by Psychological Services. Residents in the present

sample accounted for 2329 (51.4%) of the index file incidents

accounted for by residents exclusive of those residents in

the sex offender unit. The mean number of incidents appearing on

a residents index card was 6.30 (records = 369) for residents

tested by Psychological Services, and 5.410 for residents not

tested (records = 395). A single resident could have more than

one record if he was readmitted, or transferred to another


An index of the number of incidents which occurred on each

residents building during each resident's length of stay was


calculated. Index file incidents were tabulated by building and

month. Each resident's file index incidents were subtracted

from the total number of indexed incidents on their building

yielding a building incident rate which was unique to each


A total of 1367 incident reports were coded from original

Incident and Use of Force Report forms maintained security (Appendix

A). These forms were coded only for the residents in the present

study. Incidents involving building searches and those which did not

directly involve a resident were not coded.

Rates and frequency of individuals involved in different types

of incidents are presented in Table 11. These data indicated that

the frequency of individuals involved in any type of incident

(54.1%) was sufficient to expect a successful attempt at

prediction. The frequency of individuals involved in specific

types of incidents ranged from approximately 20% to 33%, somewhat

less than optimum for predictive attempts but high enough to

warrant an attempt.

An index of each building's response to incidents which were

not physically violent was created. Of the 1367 incident reports

coded, 193 were found to involve some form of verbal aggression

but no actual physical aggression toward staff, security, other

resident, self, or property. Of these verbally aggressive incidents,

70 (36.27%), resulted in physical intervention, restraint, or


Table 11

NFETC Incident Data

Incident Type: Frequency Percentage

Incidents--Any type
None 207 45.90

One or more 244 54.10

Aggressive Incidents
None 325 72.06

One or more 126 27.94

Aggressive Incidents (Extended)
None 300 66.52

One or more 151 33.48

Use of Force Incidents
None 352 78.05

One or More 99 21.95

Fighting Incidents
None 361 80.04

One or More 90 19.96

Type of Incident Mean Sample Minimum Maximum
(Monthly Rate) S. D.

Total 0.3686 0.7377 0 7.088

Aggressive 0.0794 0.1758 0 1.359

Extended 0.1173 0.2575 0 2.473

Use of Force 0.0879 0.3400 0 5.604

Fights 0.0495 0.1287 0 0.968


The rates of verbally aggressive incidents and this type of result was

tabulated by building to yield a percentage coded as "building

response" (Table 12).

Comparison of Residents Tested by MMPI and Rorschach

Since not all residents which were evaluated by Psychological

Services were administered the same battery of tests, the possibility

existed that tested residents would constitute a biased subsample of

all the residents in the study and thereby pose limits to the

generalizability of test findings.

To test the hypothesis that residents administered the Rorschach

or the MMPI were different from those who were not administered

the particular test, a series of t-tests and Chi-square analyses

were conducted.

Those residents administered the MMPI (Form-R) were compared

with the group of residents either not administered the test or

who were administered the short form MMPI-168. Data of residents for

whom the Rorschach was administered and available was compared with

data of residents for whom the Rorschach was not administered or

available. A small, undetermined number of the "no Rorschach group"

may actually have been administered a Rorschach which was not coded in

the present study because it was not scored by the examiner at the time

of evaluation.

Table 12

Building Rates of Physical Intervention, Restraint, or Seclusion
for Verbally Aggqqressive Incidents

Verbally Aggressive Physical Intervention Building
Building Incidents Restraint / Seclusion Response %

none 8 2 25.0

2 3 0 0.0

6 34 21 61.7

7 16 8 50.0

8 29 9 31.0

9 34 16 47.1

10 21 3 14.3

11 1 0 0.0

12 4 1 25.0

13 13 6 46.2

14 29 9 31.0

15 1 0 0.0

Note: Building was coded as none if it occurred on a building
other than those listed, building was not listed on incident
report, or if incident occurred before resident was actually
assigned to a building.]

Rorschach Group Differences

A comparison of those residents administered the Rorschach and

those who were not yielded the following results. All t-tests and

Chi-square analyses were performed using Microstat software (Ecosoft,


The analyses indicated no significant differences between groups

for any of the variables examined (see Tables 13 and 14). Therefore,

the null hypothesis that no between group differences exist could not

be rejected.


A comparison of residents administered Form-R of the MMPI and

those not was conducted by t-tests and Chi-square analyses. The

analyses yielded the following results.

Residents who had completed the MMPI were found to have

significantly more education (Mean = 10.282) than those not

administered the test (Mean 9.594) (Table 15).

Differences between residents administered the MMPI and those not

were found for discharge diagnoses. Chi-square analysis of Axis-I

diagnoses of "none," major affective disorder, schizoaffective

disorder, paranoid schizophrenia, and other schizophrenia found a

significant difference in the relative frequencies of these diagnoses

for those administered the Rorschach and those not (Table 16).

Proportionately more residents administered the MMPI were diagnosed as

having a major affective disorder than those not administered the test

(Table 17).

Table 13

Comparison Data for Residents Administered the Rorschach and Those Not

Mean Mean
Variable No Rorschach Rorschach F prob. d.f.














Arrest history:

total 8.068

violent 1.797

age of first 21.990

score: 80.055

Length of
Treatment: 220.14

Monthly Incident Rates:

total 0.434
aggressive 0.078

extended 0.125

use of force 0.106

fights 0.047
















.243 .6227

.117 .7326

3.233 .0729
.037 .8484
















Table 14

Results Chi-Square Comparisons of
Not Tested with Rorschach

NFETC Residents Tested and

Variable Chi-square d. f. prob. n








Abuse or


























.948 451

.107 1

Table 15

Comparison Data for Residents Administered the MMPI Form-R and Those

Mean Mean
Variable Group Group F Prob. D.F.




Arrest history:



age of first


Length of

Monthly Incident




use of force


























< .05




84.969 28.733 <.001

189.173 5.916 < .05




























Table 16

Results of Chi-Square Comparisons
Tested with MMPI Form-R

of NFFTC Rpddpnt~ Tp'~tpd and Not

Variable Chi-square d. f. prob. n








Abuse or








< .001

< .01

















.010 1

.919 451

of NFFTC Reidnt Tested d Nt...

Table 17

Chi-square Analysis of MMPI Tested Versus Not Tested Residents:
DSM-III Axis-I Discharge Diagnoses

No Major Schizo- Paranoid Other
Diagnosis Affective Affective Schiz. Schiz.


Frequency 57 5 6 107 79

Pecentage 13.57 1.19 1.19 25.48 18.81

Percentage 14.98 2.74 1.15 26.78 14.83


Frequency 47 14 2 79 24

Percentage 11.19 3.33 .48 18.81 5.71

Percentage 9.79 1.79 .75 17.50 9.69

Chi-square = 23.398, d.f. = 4, p = <.001, n = 420

Analyses also indicated that residents who were administered an

MMPI were proportionately more likely to receive a DSM-III discharge

diagnosis of some type of personality disorder other than antisocial

personality disorder (Table 18).

Residents administered the MMPI were found to have a

significantly higher measured WAIS-R I.Q. score (Mean = 84.969)

than those not tested (Mean = 77.152).

Residents administered an MMPI were found to have a lower length

of stay (Mean = 189.173 days than residents who were not administered

the test (Mean = 235.452 days, F (1,440) 5.916, p < .05.

No significant differences were found between groups for total

incident rate, aggressive incident rate, extended aggressive

incident rate, use of force, or fights (Table 15). Further analyses

(see Tables 15 and 16) failed to find significant differences.

In contrast with comparisons of resident administered and

not administered the Rorschach, significant differences were

found between residents which were or were not administered the

MMPI. Residents administered the MMPI were more educated and

achieved a higher WAIS-R IQ score upon testing.

Differences in rates of DSM-III Axis-I and personality disorder

discharge diagnoses were also found. Residents administered the

MMPI were also found to be discharged sooner than those not

administered the test. These results suggest that the group of

residents who completed the MMPI Form-R may have been functioning

better cognitively and in other ways than those residents not

administered the test.

Table 18

Chi-square Analysis of MMPI Tested Versus Not Tested Residents:
DSM-III Personality Disorder Diagnosis (Excluding Antisocial
Personality Disorder)

No Diagnosis Personality Disorder


Frequency 235 35

Percentage 52.11 7.76

Percentage 49.38 10.49


Frequency 137 44

Observed 30.38 9.76

Percentage 33.10 7.03



d. f. = 1,

with continuity correction factor = .8.886, p < .01.

without continuity correction factor = 9.655, p < .01.

n = 451


Prediction Equations

Incident data consisting of total incident rate, aggressive

incident rate, extended aggressive incident rate, use of force

rate, and fighting incident rate was analyzed by stepwise

multiple regression applied to blocks. Variables were entered in

the order of basic demographic data, diagnoses, unit assignment

and building rates, building assignment, arrest data, Rorschach

data, MMPI data, and I.Q. score (Table 3).

An F for each variable to be selected to enter the multiple

regression equations of 3.00 was selected, since it was slightly

more conservative than a F of 2.00 which was described as liberal

by Cohen and Cohen (1975). Lower values of F make it easier for

variables to enter the equation. Selection of this F value

resulted in a significance level of .05 for most variables

entering the equations.

After the multiple regression equation rates were derived, the

ability of the equations to classify residents into incident and

no-incident groups was examined. Classification rates of multiple

regression equations based on demographic data alone and on

demographic plus test data were examined by calculating the

number of residents falling above and below various cutting

scores and comparing the actual number of residents in these

categories who were actually involved in incidents.

Cutting scores were selected as to minimize the false positive

rate while maintaining a high overall classification rate. Two


separate prediction equations were computed for each type of

incident. Separate equations were computed using data exclusive

and inclusive of test data.

A summary of the resultant multiple regression equations which

were results of analyses of the demographic variables only are

presented in Table 19. These demographic results were based on

the entire sample of 451 residents. A summary of the results of

multiple regression analyses of demographic and test data are

presented in Table 20. Since not all residents completed all

tests these results were based on subsamples of residents completing

the tests.

As can be seen from these summaries, conservative estimates

of Shrunken R-squared, an estimate of the actual population

variance accounted for, are quite small or null. This is due to

the large number of predictor variables originally examined. The

liberal calculations of Shrunken R-squared, based only on the

actual variables entered after the stepwise selection procedure,

offer a much higher estimate of the actual population variance

accounted for.

Analyses were conducted for total incident rates which

included any type of disruptive incident, aggressive incidents,

extended aggressive incidents, incidents requiring the use of

force, and fighting incidents. Separate equations were derived

using only the demographic blocks of variables and those including

the test blocks. Some post hoc analyses were done which varied

Table 19

Demographic Multiple Regression Summary: Regression
Weiahts and Statistics

Overall Extended Use of
Incident Aggressive Aggressive Force Fight
Variable Rate Rate Rate Rate Rate






Big 6

Big 7

Big 10

Big 14

Big Avg
L. 0. T.

Unit II


Multi- R

Multi- R

S. Error





















































.0284 .0148

.0327 .0015


Table 19 continued

Overall Extended Use of
Incident Aggressive Aggressive Force Fight
Variable Rate Rate Rate Rate Rate

Single Variables Accounting for Largest Proportion of Variance

Variable antisoc hisp hisp hisp blg 14

R-Square .0271 .0272 .0405 .0151 .0446

Table 20

Demographic and Test Multiple Regression Summary: Regression
Weights and Statistics

Post Post
Hoc Hoc Ext
Inc Inc Aggr Aggr Aggr Fight
Variable Rate Rate Rate Rate Rate Rate


.1072 .1336

-.0649 -.0761 -.0563

.0899 .0932


.0018 .0021







B1g 14

Sum C

path %


M:C (100)



Multi- R

Multi- R

S. Error







.1053 .0710

.0000 .0000 .0000








































.0000 .0000

Table 20 continued

Post Post
Hoc Hoc Ext
Inc Inc Aggr Aggr Aggr Fight
Variable Rate Rate Rate Rate Rate Rate

Single Variables Accounting for Largest Proportion of Variance

Variable antisoc antisoc hisp hisp hisp blg 14

R-Square .0413 .0402 .0396 .0396 .0345 .0397


the ofder of entry of test blocks, and used additional Rorschach


Total Incident Rates

Results of the multiple regression procedure conducted on overall

incident rates are presented in Table 21. Results of the analysis

prior to entry of test data indicated that each of the variables age,

Hispanic ethnic group, antisocial personality disorder, other Axis-I

diagnosis, and building 14 assignment contributed significantly when

added to the equation.

The resultant multiple regression equation accounted for

10.23% of the incident rate variance, and was significant at the .001

level. Young Hispanic males with a diagnosis of antisocial personality

disorder had the highest predicted incident rates. The diagnosis of

antisocial personality disorder was the single variable which accounted

for the largest proportion of incident rate variance (4.30%), and

was the variable which contributed the highest proportion of

independent variance in the prediction equation.

Classification rates based upon the resultant equation of

predicted rate = -.00816 age + .39952 Hispanic ethnic group +

.97753 anticosial personality disorder + -.26827 other Axis-I

diagnosis + .21660 building 14 assignment + .48961 are presented

in Table 22.

Table 21

Incident Rate Blockwise Multiple Regression Results

Block 1. Demographic Variab

Step Variable D. F.

1. Hispanic 1,441

2. divorced

3. age



Regression Block One results

Variable Regression

age -.00652

Hispanic .34527

divorce -.14876

constant .54735















< .05






R Squared








Multi R




Partial r^2











Analysis of


11.18729 3

232.97605 439

244.16334 442









Full Text
xml version 1.0 encoding UTF-8
REPORT xmlns http:www.fcla.edudlsmddaitss xmlns:xsi http:www.w3.org2001XMLSchema-instance xsi:schemaLocation http:www.fcla.edudlsmddaitssdaitssReport.xsd
INGEST IEID ENTM6EF8O_0TWUD4 INGEST_TIME 2012-02-20T22:40:28Z PACKAGE AA00009084_00001