Citation
Prediction of dangerous behavior for release of the criminally insane in Colorado

Material Information

Title:
Prediction of dangerous behavior for release of the criminally insane in Colorado
Creator:
Koppin, Mary Katherine
Publication Date:
Language:
English
Physical Description:
vii, 106 leaves : illustrations ; 28 cm

Subjects

Subjects / Keywords:
Insane, Criminal and dangerous -- Colorado ( lcsh )
Genre:
bibliography ( marcgt )
theses ( marcgt )
non-fiction ( marcgt )

Notes

Bibliography:
Includes bibliographical references (leaves 99-106).
General Note:
Submitted in partial fulfillment of the requirements for the degree, Doctor of Philosophy, Graduate School of Public Affairs.
General Note:
School of Public Affairs
Statement of Responsibility:
by Mary Katherine Koppin.

Record Information

Source Institution:
University of Colorado Denver
Holding Location:
Auraria Library
Rights Management:
All applicable rights reserved by the source institution and holding location.
Resource Identifier:
22692892 ( OCLC )
ocm22692892
Classification:
LD1190.P86 1990d .K66 ( lcc )

Downloads

This item has the following downloads:


Full Text
PREDICTION OF DANGEROUS BEHAVIOR FOR RELEASE
OF THE CRIMINALLY INSANE IN COLORADO
by
Mary Katherine Koppin
B.S., Southern Colorado State College, 1970
M.A., University of Colorado, 1974
A thesis submitted to the
Faculty of the Graduate School of the
University of Colorado in partial fulfillment
of the requirements for the degree of
Doctor of Philosophy
Graduate School of Public Affairs
1990


This thesis for the Doctor of Philosophy
degree by
Mary Katherine Koppin
has been approved for the
Graduate School
of Public Affairs
by
Eric D. Poole
(7
1 over Burton
Date
Richard Pasewark


Koppin, Mary Katherine (Ph.D., Public Administration)
Prediction of Dangerous Behavior for Release of the Criminally
Insane in Colorado
Thesis directed by Professor Eric D. Poole
ABSTRACT
The public policy of preventive detention of the mentally ill is
legitimized by insanity statutes in Colorado. Not Guilty by Reason of
Insanity (NGRI) cases were followed after conditional release from
the states only psychiatric security hospital. Agency release
recommendations, made on a danger products test, were evaluated
for predictive accuracy by Relative Improvement Over Chance
(RIOC). Disruptions before and during hospital career were tested
for predictive relationships with criminal recidivism, rehospitaliza-
tion, and revocation of conditional release. Discriminant analysis
was used to determine which variables were useful in predicting
membership in risk and outcome groups. Clinical predictions were
found to distinguish NGRIs who were subsequently dangerous. Crime
of commitment, youth, and length of stay were not predictive of
dangerous outcomes. Progression through a treatment program of
gradual reduction in external controls was correlated with clinical
predictions and with outcome. Conformity to the program resulted
in favorable recommendations for release and fewer dangerous
behaviors after release. Diagnostic and program differences within
the sample of 109 NGRIs were suggested to reflect two distinct
populations.
Signed
I _ thesis


CONTENTS
CHAPTER
1.INTRODUCTION .................................................. 1
The Research Problem ....................................... 2
Comparability of Samples .............................. 3
Replication of Methods ................................ 4
Generalizing from the Findings ........................ 5
The Public Policy Problem ............................... 6
Legal Background....................................... 6
Clinical Background .................................. 13
Methodological Background ............................ 14
Scope of the Problem.................................. 16
Purpose of the Present Study.......................... 19
Organization of the Dissertation............................. 19
2.REVIEW OF THE RELEVANT LITERATURE .............................. 22
Legal Foundations............................................ 23
Jones v. United States................................ 24
Legal Definitions of Danger........................... 26
Outcome Criteria for Danger .......................... 28
Psychological Assessment of Danger .......................... 30
Personality Scales ................................... 31
The Bridgewater Legacy .................................... 33
Psychopathy and Validity of Diagnosis ................ 34
Prediction Accuracy .................................. 36


The Baxstrom and Dixon Cases
37
The Legal Dangerousness Scale ......................... 39
Political Predictions and Dixon Cases ................. 41
Prediction Accuracy ................................... 42
Research Propositions ...................................... 43
3. METHODS....................................................... 45
Design and Prediction Models ............................... 46
Selection of the Sample..................................... 48
The Treatment Setting ...................................... 49
Program Description ................................... 50
Prerelease Testing in the Community ................... 52
Maximum Security Programs ............................. 53
Description of the Sample................................... 53
Demographic Characteristics ........................... 54
Length of Hospital Stay and Crime...................... 55
Release Procedures ......................................... 56
The Legal Release Criteria ............................ 56
Patient Petition for Release . ,................... 57
Psychiatric Recommendations ........................... 58
The Administrative Review ............................. 58
The Official Agency Recommendation .................... 59
Adversarial Release Hearings .......................... 60
Data Sources and Limitations................................ 61
Patient Data System.................................... 62
Hospital Records ...................................... 62
Level of Measurement................................... 63
v


Validity and Reliability
63
Operational Definitions .................................... 64
Disruption After Release .............................. 65
Disposition After Disruption .......................... 66
Prehospital Disruptions ............................... 66
In-Hospital Disruptions ............................... 68
Statistics........................................... 69
4. RESULTS................................................ 73
Definition of Dangerous Disruption ..................... 73
Statistical Analyses ....................................... 77
Dangerous Outcomes by RIOC............................. 78
Agency Opposition by RIOC.............................. 82
Discriminant Function Analysis of Dangerous Outcomes 83
Discriminant Function Analyses of Opposition .... 85
Summary of Findings.................................. 86
5. DISCUSSION ................................................... 89
Base Rates and Probability Models .......... 91
Explicit and Replicable Criteria ...................... 92
Increased Degree and Probability of Harm............... 93
Research Relevant Evidence ............................ 94
Programs to Reduce Propensity for Violence ............ 95
Provision of Thorough Review of Confinement .... 96
BIBLIOGRAPHY.............................................. 99
vi


TABLES
3.1. Crime of NGRI Commitment by Length of Hospital Stay
in Months: Conditional Releases 1980-1986. (n=109) . . 55
4.1. Disruptions Involving Arrest or Rehospitalization
after Conditional Release (n = 109)..................... 74
4.2. Disposition of Disruptive Events Involving Arrest or
Rehospitalization after Conditional Release (n = 109). 75
4.3. Types of Disruptions Involving Arrest or
Rehospitalization after Conditional Release
by Hospital Prediction (n=109).......................... 76
4.4. Accuracy of Predictions of Dangerous Behavior
After Conditional Release (n=109)....................... 79
4.5. Hospital Recommendations For or Against Release by
In-Hospital Disruptions as Predictors (n=109)........... 80
4.6. Predicted and Actual Group Membership for Dangerous
Outcomes with Equal and Bayesian Probability Models. . 84
4.7. Predicted and Actual Group Membership for Hospital
Opposition with Equal and Bayesian Probability Models. . 86
vii


CHAPTER 1
INTRODUCTION
The public policy of preventive detention of the criminally
insane is based on an assumption that psychiatrists and other mental
health professionals can distinguish dangerous from safe persons.
There are few empirical investigations of danger prediction, and most
researchers concluded that psychiatric opinions about future danger-
ousness were usually wrong (Ennis & Litwack, 1974; Monahan, 1984;
Steadman & Cocozza, 1974; Thornberry & Jacoby, 1979).
The lay public believed psychiatrists could predict violence; the
research findings suggested they could not; but the law required that
they must predict. Psychologist and attorney Stephen Morse (1983)
recognized the legal mandate and recommended the use of expert
evidence with improved social science:
1. to make the criteria more explicit,
2. to increase the degree and probability of harm
required,
3. to admit only that evidence that research demon-
strates is relevant.
4. to offer programs to those confined that might
reduce their propensity for violence,
5. and to provide thorough periodic review of the
continuing need for confinement, (p. 18)
Morses objectives were formulated in light of the consistent finding
that clinical predictions were accurate in only one third of the cases.


Statistical predictions achieved better accuracy, but relied on
arbitrary social predictors (age, sex, race).
The literature on the prediction of danger is primarily the work
of a small number of public policy critics who reported inaccurate
risk assignment by psychiatrists. The direction of error was consis-
tently reported as one of overclassifying mentally ill offenders as
dangerous, whether by clinical or statistical assignment of risk (Pfhol,
1978; Steadman & Morrissey, 1981). The infrequency of violent
events was one reason for low accuracy rates. Rare events and
random probability were considered in the predictive analyses of
Loeber and Dishion (1983) in Oregon delinquency studies, a statistic
applied here to an adult sample of criminally insane patients.
Most investigators measured prediction accuracy in retrospective
studies by classifying certain outcome behaviors as dangerous, and
then searching for offender characteristics that might have been
predictive of outcome. Arrest for crimes of violence and/or rehos-
pitalization for assaultive behavior were defined as dangerous
recidivism in nearly all of these core studies.
The Research Problem
Because most offenders, including those who are criminally
insane, remain confined when classified as dangerous, researchers
seldom have opportunity to test the accuracy of predictions.
Dangerous mental patients are detained in prisons or hospitals
without the freedom to test their danger classification. Courts
interrupted such continued confinement when they ordered the
2


release of several hundred patients in maximum security hospitals of
New York in Baxstrom Herold (1966) and Pennsylvania in Dixon v^
Commonwealth (1971). Due process rulings provided the first
opportunities for follow-up of presumably dangerous and insane
criminals. The courts action was acclaimed by social scientists as
an opportunity for a natural experiment, a study of the public policy
of the preventive detention of mentally ill offenders.
Comparability of Samples
Analysts broadly defined the Baxstrom and Dixon populations as
criminally insane, generalizing from mentally ill prisoners, pretrial
incompetents, sexual psychopaths, and patients unmanageable in civil
hospitals. Comparability among research samples and distinctions
between mental health and correctional populations were initial
methodological problems noted by researchers. Baxstrom and Dixon
release samples were primarily prison inmates who became ill or
troublesome and were transferred to security hospitals. Neither
sample was specifically predicted to be dangerous (Greenland, 1985)
nor had replicable variables been used to classify the patients as
dangerous at follow-up.
The technical definition of criminal insanity includes only
criminal defendants who have been acquitted of a crime by being
found Not Guilty by Season of Insanity (NGRI). NGRI defendants
have been relieved of criminal responsibility by avoiding culpability
and accountability before the law, unlike the convicted criminal who
may be punished. The acquittal makes any subsequent criminal
sanctions for detention of the NGRI conceptually quite different
3


from sanctions applied to convicted and sentenced correctional
populations. Comparison of samples, replication of methods, and
generalization of findings are as relevant to the research problem as
the theoretical issues of public policy and preventive detention.
Replication of Methods
Dangerousness is often assumed by clinicians to be a charac-
teristic or a trait within the individual, a propensity that can be
diagnosed. No adequate typology of violent persons exists, however,
according to Mulvey and Lidz (1985). No psychological tests directly
measure such a propensity, although indirect measures of hostility,
aggression, and psychopathy have shown promise in Canadian studies
(Hart, Kropp, & Hare, 1988). Persons who scored high on the HARE
Checklist of Psychopathy (PCL) had higher rates of criminal recidiv-
ism and revocation of conditional release than nonpsychopaths and
other lower scoring subjects. It is not known, however, how much
dangerous behavior can be accounted for by psychopaths. The
following work includes a sample of chronically mentally ill (the
majority were nonpsychopathic) patients. Base rates of psychopathy
within samples of hospitalized offenders may differ among jurisdic-
tions according to commitment laws.
Prediction of danger studies, typically retrospective and exp-
loratory, have generally not used instruments that have been norraed
or validated on other samples. New York scholars Cocozza and
Steadman (1974) did produce a four item "scale" which was a summary
of criminal history in binary code. The New York Legal Dangerous-
ness Scale yielded strikingly similar results on a sample of Dixon
4


patients, and with a sample of younger and less institutionalized
NGRI patients released in Colorado (Koppin, 1977). Predictive
accuracy, base rates of recidivism, and criminal histories were also
quite similar in Gordons (1982) comparisons of Maryland patients
with New York, Massachusetts, and Colorado samples. The primary
problem with pre-commitment variables of criminal record as predic-
tors is their fixed and permanent nature. Only growing older will
take an offender out of the high risk group. Nothing in treatment
can change his criminal history as fixed predictors of danger.
Generalizing from the Findings
The reported findings of inaccuracy and over prediction of
danger led to protectionist criticisms in the literature. Conserva-
tives charged libertarians with generalizing from an opportunistic
case mix, aged populations, convicted inmates and patients who were
never charged with crimes, and imprecise definitions. Spokesman
Saleera Shah of The National Institute of Mental Health led public
policy analysis with his proposals for a research paradigm (1975,
1978, 1981, 1986). He recommended distinctions for the major
concepts of "dangerous" behavior, violence, and "dangerousness."
The terms had various uses and definitions in the law and in the
psychological literature.
The early studies formed what Monahan (1984) called the first
generation of prediction research with mentally ill criminals. The
works defined the state of the art of psychiatric prediction and the
empirical study of relevant populations. Forensic psychiatrists and
psychologists were following the Baxstrom and Dixon results, along
5


with the clinical predictions reported by Kozel, Boucher, and Garofalo
(1972) on a sample of sexual psychopaths from a Massachusetts
security hospital. The Kozel findings supported a treatment model
and were more acceptable to clinicians than the New York and
Pennsylvania reports of psychiatric inability to diagnose and predict.
Considerable influence on public policy was attributed to the early
studies, with technical criticisms receiving less attention in the policy
arena (Klassen & OConnor, 1988).
Conclusions that clinical predictions were most often wrong
may have been generalized from insufficient evidence and faulty
methods (Floud & Young, 1982; Gordon, 1977, 1982; Mulvey & Lidz,
1985). Illusory correlations and problems of colinearity were alleged
by these and other authors. Experts in probability theory said that
not enough information was known about base rates, or the frequency
of dangerous behavior among particular populations, to have developed
sophisticated probability theory or models. Because most items
reported to be associated with dangerous behavior were at a categor-
ical (nominal) level of measurement, parametric statistics were rarely
appropriate.
The Public Policy Problem
Libertarian scholars concluded that society locked up far too
many innocent persons under the guise of danger predictions.
Inaccurate classification of persons as dangerous who are actually
safe leads to costly and unnecessary utilization of hospital and
prison resources, according to the libertarian perspective. In a
6


medical or clinical model, the false positive is judging a healthy
person to be sick and proceeding with a treatment. This is consid-
ered to be preferable to the false negative of misdiagnosing health
and failing to treat. The policy issue for civil libertarians is one of
false positives and unnecessary treatment and confinement.
Monahan (1973, 1981) was representative of social theorists who
used primary data generated by Baxstrom and Dixon cases in public
policy debate. Policy analysts monitored reports on the 1,000 court
ordered releases from New York and Pennsylvania security hospitals.
Findings were that criminal patients were not very dangerous and
behaved much like institutionalized civil patients in community
placements (Steadman & Cocozza, 1974; Steadman & Keveles, 1972;
Thornberry & Jacoby, 1979).
Conservative practitioners (Deitz, 1979; Floud & Young, 1982;
Kozel et al., 1972) argued in favor of social control and public
protectionist policies. Psychiatrists using a danger prediction model
are said to make conservative predictions and retain for treatment,
rather than release the patient and risk a false negative. When a
person is predicted dangerous and is simply confined for treatment,
there is less political and media impact than when a person who was
predicted to be safe commits a dangerous act.
Viewed from the perspective of public protectionists the policy
issue is one of false negatives. The public administrators at the
study hospital for this research described their mission as protec-
tionist with conservative policies to minimize false negatives (see
Appendix). The avoidance of critical media coverage is also implied
7


from policies of the Governor's office. Erroneous prediction of
safety may be followed by hasty legislation passed in response to a
violent act by a former mental patient. Subsequent restrictive
changes in public policy may proceed in the absence of rational
information (Mulvey & Hilz, 1985).
An emphasis on civil liberties leads to a focus on the error of
over prediction of danger, whereas an emphasis on public protection
leads to a focus on the error of inaccurate prediction of safety.
The right to treatment in the least restrictive environment, as well
as constitutional rights to due process, are premises from the
libertarian perspective. The responsibility to protect the public, and
to avoid liability for the dangerous behavior of psychiatric patients,
are premises from the protectionist perspective.
Legal Background
Although there was little scientific evidence that mental health
professionals could accurately predict individual events of dangerous
behavior, federal and Colorado laws continued to require release
decisions based on psychiatric assessment of the likelihood of
dangerous behavior. Eligibility for release from an NGRI commitment
in Colorado (CRS 1983, 16-8-120) requires a psychiatric opinion that
the patient "no longer suffers from an abnormal mental condition that
is likely to cause him to be dangerous to himself, to others, or to
the community in the reasonably foreseeable future."
The problem of long-term prediction of danger for the criminally
insane contrasts with the short-term (civil) assessment of imminent
danger. The legal criterion for conditional release required the
8


psychiatrist to relate abnormal mental conditions to a propensity for
future dangerous behavior. The statutes do not provide a definition
for foreseeable future; in clinical practice the treatment plan usually
required a minimum of one year on conditional release. In the NGRI
sample it was not unusual for psychiatrists to report that the
schizophrenic would remain not dangerous if he continued neuroleptic
medication, or that the alcoholic might again become dangerous if he
were to drink. Psychiatric assessment and testimony in criminal
insanity cases thus takes the form of clinical prediction. Clinicians
recommended restrictions on gun ownership, use of substances, and
compliance with prescribed medications, which were used as court
ordered conditions for remaining on release status.
Actuarial or statistical predictions are seldom used in decisions
about individuals. Scales or instruments are more often a research
tool for retrospectively applying measures to groups of offenders
after clinical predictions are implemented and follow-up data are
obtained. Actual application of scales or measures that have no
norms or validity would be inconsistent with civil rights legislation,
and with the ethics of The American Psychological Associations
standards for tests and measurements (1985).
Statistical associations with dangerous behavior have included
demographic items of youth, sex, race, juvenile record, or number of
arrests and incarcerations. Actuarial models do not rely on causal
theories, but items are used to show that increased probability of
dangerous behavior occurs with increased risk factors that assign
offenders to risk prediction groups. An ideal situation for court
9


work would be to state an individual's probability as a member of a
particular high (or low) risk group, recognizing the probability as
only a base rate for some known group that may be similar to the
individual.
The Colorado release test is a type of "product" test. Neverthe-
less, clinical practice observes dangerous persons who are not
mentally ill, and many mentally ill persons who are never dangerous.
Causality (that is, the abnormal mental condition causes the danger-
ous act) is implied in a number of commitment statutes. In the New
Hampshire insanity test criminal behavior is considered the "product"
of mental disease or defect (Keilitz & Fulton, 1984). Disturbed or
criminal behaviors were thought to be the result of hallucinations,
delusions, or disordered emotions that came to be legally defined as
mental disease or defect. The United States Courts used a product
test for a generation, associated with the case of Durham v. United
States, 1954). The Durham product test was cumbersome, alleged to
encourage wide psychiatric speculation, and was abandoned in favor
of the American Law Institutes (ALI) Model Penal Code.
The ALI standard was a test of "substantial capacity to ap-
preciate the criminality of his conduct or to conform his conduct to
the requirements of law" (Keilitz & Fulton, 1984). The legal
concept of capacity was an insanity test more tolerable to psychia-
trists than the concept that implied causality. Not until legal
reforms in 1983 (and only for NGRIs whose crimes were commited
after July 1, 1983) did Colorado adopt language similar to ALI. The
legislature retained the danger criterion, similar to a product test,
10


and added the capacity to appreciate the law in the release test.
The burden of proof was on the patient to prove safety and sanity.
The Colorado release test for this sample required a psychiatric
assessment of an individuals likelihood of dangerousness, as the
result of an abnormal mental condition. Dangerous outcomes among
109 conditional releases reported in Chapter 4 were categorized
according to behaviors handled by police and correctional authorities,
and behaviors that required psychiatric hospitalization. These and
other definitional problems with dependent variables were detailed in
the review of the relevant clinical and legal literature.
Colorado law assumes that psychiatric inpatient hospitalization
is necessary for an unspecified period of time for all defendants
aquitted by reason of criminal insanity. The issue of dangerousness,
however, is not legally relevant at the time that a Colorado
defendant is tried and adjudicated insane. The finding of NGRI for
the sample in this study was based on each defendants "cognitive"
ability to know right and wrong, or an inability to refrain from the
wrong (the 1843 MNaghten test and the irresistible impulse).
Colorado law interprets the MNaghten rule by defining insanity as a
defect in reason (that is, cognitive ability) due to disease or defect
of mind at the time of a criminal act (CRS 1973 16-8-101). The
affirmative defense assumes the defendant coramited the act and
presumes continued insanity.
Psychopathy (antisocial or sociopathic disorders) and substance
abuse were not excluded during the study period by legal definitions
of mental disease or defect. Consequently, the sample included some
11


patients who resembled correctional cases and others who resembled
the chronically mentally ill. The appropriateness of the insanity plea,
or the many philosophical forensic arguments about criminal respon-
sibility, are not particularly relevant to the subject of prediction of
dangerous behavior of NGRI patients after hospitalization. One case
study approach, for the reader interested in the philosophical
controversies, is The Trial of John VL Hinckley, Jr. by Low, Jeffries,
and Bonnie (1986). Portions of the trial transcripts provide an
excellent introduction to diagnostic issues.
An NGRI finding is an automatic and indeterminate commitment
to a security hospital rather than prison, and the defendant can
never be tried again for the crime. The U.S. Supreme Court affirmed
the principle of indefinite duration of hospitalization for the crim-
inally insane by their decision in Jones v^ United States (1983).
Michael Jones, a shoplifter, spent many years in Saint Elizabeths, a
federal security hospital. In response to an equal protection argu-
ment, the court ruled that Jones belonged to a special class of NGRI
persons, unlike mentally ill persons under civil commitment.
Jones was contesting his indeterminate length of stay. The
danger criterion for release was not specifically addressed by the
defendant, prosecutors, or the court. The court did rule, however,
on the constitutionality of involuntary and indeterminate confinement
in a security hospital for this class of patients. No matter how
trivial or serious the instant offense, hospitalization did not con-
stitute punishment and could continue for Jones and several thousand
12


other NGRIs in United States jurisdictions. The spirit of preventive
detention was upheld, if not the precise standard of predicted
danger.
Clinical Background
Colorado insanity laws allow a rehabilitation and treatment
model. The courts approve gradual increase of patient privileges and
reduction of supervision and external controls. This model is
supposed to test patient dangerousness, and prepare him for re-
entry in the community. Patients in the present research were
treated in a graduated security level program that provided privilege
incentives for progression through certain treatment phases. The
taking away of privileges by regressive transfer to a higher security
level was hypothesized to be one of several indices of the patients
failure to demonstrate safety by "going through the system."
The present work, unlike most of the previous prediction
studies, did include variables that might infer treatment effects.
The present sample came from two distinctly different programs of
rehabilitation, organized around patient characteristics, generally
categorized as level of social functioning. Recent Massachusetts
work with a typology of sex offenders included the construct of
social functioning (Rosenberg et al 1988). Their classification
scheme was similar to the treatment tracking of patients in this
sample, diagnosed and assessed according to their premorbid lifesyles.
Treatment Track I patients tended to have been psychotic,
chronically ill, socially isolated with few verbal skills, prosocial in
values, and in need of social learning of daily living skills in a
13


supportive atmosphere. Track II patients tended to be personality
disorders or substance abusers, socially gregarious and verbal,
antisocial in values, and in need of character development in a
group and community milieu. Program patients were expected to
have distinguishable differences in characteristics, treatment
disruptions, predictions and dangerous outcomes.
Methodological Background
Retrospective study of variables that might have had predictive
value in most of the core studies found that juvenile record, prior
violent crimes, severity of the instant offense, and age were most
frequently and significantly associated with subsequent dangerous
behavior. Criterion behaviors were assaults and rehospitalization
and/or rearrest for serious crimes of violence. A serious problem,
however, in practical application of these predictor variables is that
risk level is assigned at the time of commitment (record, offense, and
age) and is fixed except for growing older. The New York and
Pennsylvania samples were each found to be safer if over 50 years of
age. If crime is a function of youth and masculinity, then todays
generation of offenders would have to be incarcerated for decades.
The research here concerned preventive confinement of the
criminally insane on the basis of predictions of danger, and concerned
the methods and models for analyses of those predictions. Colorado
case law reflected a broad definition of dangerousness (a likelihood to
become dangerous to oneself, others, or the community) in the
foreseeable future. Such a broad definition would encompass all
subsequent illegal activity, rather than only seriously assaultive
14


crimes. A major part of the present analyses focused on the
inherent methodological problem of safe/danger prediction dichotomies
and safe/danger outcome classifications. The implied zero-one
probabilities led to a number of experimental groupings of predictor
variables and classifications of behaviors after release.
Disruptions after release were classified! as in previous studies,
according to the severity of criminal charges, and the action of the
courts of commitment in revocations of conditional release. Arrest
for any reason and/or rehospitalization under any legal status was
defined as disruption. Reasons were then classified according to
major and minor disruptions, deducing the subgroup defined as
dangerous (n=22) according to the frequencies and distributions of a
number of disruptive behaviors. Analyses were described in Chapters
3 and 4, and results were interpreted in Chapter 5.
Equal probability and actual probability distributions were
derived by discriminant function analyses. Initial analyses began with
many potential predictor variables (37) and a goal of variable
reduction for the most parsimonious explanations. Items were tested
for significant discrimination of dangerous outcomes, and the same
variables were tested for differentiating groups defined by hospital
predictions of danger.
Bayesian probability models consistent with a civil libertarian
approach were applied to minimize erroneous predictions of danger.
Similar to the contrast made by Steadman and Morrissey (1981), an
equal probability model consistent with a public protectionist ap-
proach was compared to minimize erroneous predictions of safety.
15


Discriminant function analyses were used to reduce the number of
variables that distinguished dangerous from safe outcome groups, and
groups opposed by the hospital for release.
Studies with primary data on mentally ill criminal offenders
provided the methodological model for the secondary literature and
for this research. Except for recent RAND studies which used low,
moderate, and high probability group assignments with correctional
samples (Conrad, 1985; Greenwood & Turner, 1907), most researchers
used a zero-one probability assumption to make retrospective assign-
ments to dangerous or not dangerous risk categories. Those assign-
ments implied either a 100 percent probability or a zero probability
to an individuals propensity for danger. Outcome variables of
recidivism or relapse were similarly dichotomized. Nearly all studies
used this method to report unacceptably high rates of false positives.
The zero-one assumptions and previous findings were challenged in
the theoretical work of Floud and Young (1982), Gordon (1977, 1982),
and Holland, Levi, and Beckett (1981), discussed here in Chapter 5.
Scope of the Problem
Shah (1978) estimated that 50,000 decisions were made daily in
the United States on the basis of offender potential for danger. A
daily population of about 80,000 to 100,000 persons were involuntary
commitments in mental hospitals (Morse, 1983). Prison security clas-
sification, bail and bond eligibility, civil hospitalization, probation,
parole, and release from insanity commitment are examples of
decisions that are essentially clinical. Common to most such judg-
ments is consideration of the individual offenders criminal record.
16


When a psychiatric record or violent behavior was also present in the
history of California parolees( the likelihood of predicting future
dangerous behavior was greater (Wenk et al., 1972). The Wenk
studies were the first with large samples and multivariate analyses,
hut the false positive ratio was quite high.
The magnitude of the policy and research problems is affected
by the public perception that links violent crime and mental illness.
The spectacular nature of violence, as well as media reporting of a
few notorious cases, as in The Denver Post on December 18, 1982,
creates strong public opinion about criminal insanity as a defense.
The public is disparaging of mental health and criminal justice
systems that "allow crazy criminals to walk free." Pasewark,
Jeffrey, and Bieber (1987) reported, however, that the NGRI plea
was entered in less than one percent of all Colorado criminal cases,
and was successful in less than half of those. Although small in
absolute numbers, the commitment and release of the criminally
insane are perceived to be a public policy problem, one which
warrants empirical investigation.
In addition to the broader background of research findings,
clinical issues, and legal foundations, there was a decidedly political
background for public administration and agencies of social control.
A political and public policy crisis contributed to the timeliness and
appropriateness of this particular study of dangerous behavior among
the Colorado insane. John Bromley (1980) was one of many area
journalists who indicted police and mental health professionals for
not predicting and preventing a rash of homicides in the Denver area.
17


A number of clients known to mental health agencies were respon-
sible for over a dozen murders within a few months of each other.
An Executive Order by Governor Lamm (April 14, 1900) called for
study of the publicized cases and the scientific analyses of relevant
target populations for the prevention of violence.
A legislative task force recommended reform of conditional
release laws for NGRI commitments, even though the Denver crimes
had not involved any former NGRIs or conditional releases. Senate
Bill 1, sponsored by Senator Ruth Stockton (R-Lakewood), was
passed in the 1981 Session just after the John Hinckley presidential
assasination attempt. A Colorado psychiatrist was sued by survivors
of the Hinckley shooting for not predicting and preventing his
patient's violent behavior. Hinckleys subsequent insanity trial
provoked massive legal reform in most states and the federal
jurisdictions. These events formed the media background for policy
debate.
Parallel to the executive's problems in mental health agencies
was a class action suit by inmates of the state prisons in Ramos et
al. v^ Lamm et al. (1979). Classification schemes, crowding, psycho-
logical services, and administrative procedures were scrutinized. The
federal judge concluded that the old maximum security unit was unfit
for human habitation. Judge Kanes 75 page order included ten pages
about inadequate health care, with direct criticism of the state
hospitals forensic unit. Hospital decisions were influenced by this
background of conservative and protectionist policy debate.
Purpose of the Present Study
10


Objective, valid, and replicable methods for assigning individual
risk of future danger in forensic populations have yet to be devel-
oped. This work retrospectively evaluated agency predictions made
for the conditional release of 109 male NGRIs in a six year release
cohort. Accuracy of clinical predictions was measured by follow up
for criminal recidivism or psychiatric rehospitalization as a result of
dangerous behavior in the community. Attempts to type the kinds of
disruptive behaviors according to severity were similar to the Holland
et al. studies and a 15 year longitudinal study of conditionally
released patients in Maryland (Spodak et al., 198A).
Findings were expected to contribute to a sparse NGRI data
base and to provide clinicians with objective feedback about danger
assessments. There are few, if any, studies with primary data on
danger predictions made for an NGRI sample of younger, less
institutionalized patients. Unlike previous studies, specific assess-
ments of likelihood for danger were made by psychiatrists on a legal
criterion. Technical methods for evaluating the accuracy of predict-
ions were explored. Measures for comparing predictive value to
results achieved by chance distribution were taken from delinquency
research (Loeber & Dishion, 1983). The evaluation models were
asserted to be as important a contribution to clinical prediction as
the variables tested by those models.
Organization of the Dissertation
The first chapter introduced the subject matter of clinical
prediction of dangerous behavior as an issue of public policy. The
research problem included the need for comparable samples, replicable
19


methods, and validity for generalizing findings. The theoretical
perspectives of civil libertarians and public protectionists provided a
background for understanding some practical issues of preventive
detention and inaccurate predictions. An overview of the research
and methodological difficulties in policy analysis specified the major
limitations.
Variables reported to be of predictive value were taken from
the literature on psychiatric predictions of dangerous behavior
reviewed in Chapter 2. The theoretical model was provided by the
assumptions and conclusions of previous empirical analyses and the
secondary literature generated by policy analysts. Study propositions
were derived from the critique of the literature and clinical practices
at the public hospital for NGRIs studied here. Hypotheses were
stated at the end of Chapter 2, and were approved in colloquium in
September of 1987.
Data sources included hospital and court records and federal
and state arrest records. Identification of the sample and descrip-
tive items of criminal and psychiatric history came from the hos-
pital's Forensic Patient Data System, a computerized data base for
research and management information. Predictor variables used to
test the propositions were listed and defined in Chapter 3, Methods
and Design. A description of previously reported probability models
was developed to explain the notions of false positives and civil
libertarian theory. The problem of false negatives was incorporated
in the description to explain the public protectionist perspective.
The Methods chapter described the calculation of a measure of
20


the predictive items' Relative Improvement Over Chance (RIOC). The
concept represents a slightly different approach in the forensic
literature to assessing the value of predictive variables. Discrim-
inant function analysis was chosen to test combinations of predictor
variables because of the categorical nature of the outcome groups.
Issues of appropriate statistical design for the levels of
variable measurement were discussed as one of the data limitations.
Selected summary tables reported study findings and statistical
analyses in Chapter 4. Various methods of assigning risk level were
used retrospectively, and actual rates of dangerous behaviors were
analyzed to determine appropriateness of risk assignments made by
agency predictors and experimental variables. A model of analyzing
false positives, false negatives, and total accuracy rates is in keeping
with the previous methods of reporting clinical and statistical
predictions. One purpose of additional research was to determine if
knowledge about disruptions in patient histories and hospital careers
would yield any better accuracy for prediction than chance alone.
Gordons (1977, 1982) alternative analyses with probability theory
and his criticisms of assumptions used in most primary data analyses
provide a framework for the discussion of findings in Chapter 5.
Implications for clinicians and their programs for the criminally
insane are drawn from the findings. The analyses conclude with
recommendations for including events during hospitalization, as well
as traditional fixed variables of age and criminal record, in assigning
risk for future danger.
21


CHAPTER 2
REVIEW OF THE RELEVANT LITERATURE
Case law and legislative reform during the study period provided
the pertinent legal foundations for understanding the research models
reviewed here. United States Supreme Court decisions in 1983
referenced the first-generation core studies of Baxtrom and Dixon
patients, and relevant civil case law regarding dangerousness.
Legal precedents in criminal law were distinguished by policy analyst
Shah (1978) as "judiciary definitions" of danger. The diagnostic
model of prediction by psychodynamic formulations (Kozel, 1982;
Kozel et al., 1972) provided the pertinent clinical foundations for
understanding the theoretical background.
Technical approaches to measuring prediction accuracy became
more sophisticated in the clinical and correctional literature a
generation after the core studies. Monahan (1984) chided the
American Civil Liberties Union for concluding a 95 percent error
rate for clinical prediction, citing only five studies.
Rarely has research been so uncritically accepted and so
facilely generalized by both mental health professionals
and lawyers as was this first-generation research on the
prediction of violence. The careful qualifications the
researchers placed on their findings and the circumscribed
nature of the situations to which they might apply were
forgotten in the rush to frame a bumper-sticker conclusion
psychiatrists and psychologists cant predict violence
and paste it on every policy vehicle in sight, (p. 10)


The review here examined the original core studies to show the
tedious development of a body of knowledge from the research.
Legal Foundations
A review of relevant case law revealed how courts have applied
diagnostic concepts to legal definitions and release decisions. The
standards of proof (a mere preponderance of evidence, clear and
convincing evidence, and evidence beyond a reasonable doubt) serve
to allocate the risks of error among litigants. The standards were
said by Reisner (1985) to be a function of the severity of conse-
quences to society of an erroneous decision. The legal question is
one of balancing the social costs of individual liberty against public
safety. In Addington v^ Texas (1979) the plaintiff argued against a
preponderance standard for civil commitment:
The individual should not be asked to share equally with
society the risk of error when the possible injury to the
individual is significantly greater than any possible harm
to the state. We conclude that the individuals interest
in the outcome of a civil commitment proceeding is of
such weight and gravity that due process requires the
state to justify confinement by proof more substantial
than a mere preponderance of the evidence. (Reisner, 1985,
p. 413)
Addingtons argument that the state should have to prove his need
for commitment beyond a reasonable doubt was rejected. The
standard of clear and convincing evidence was ruled by the U. S.
Supreme Court. The Court rejected the more stringent standard
because of the "subtleties and uncertainties in psychiatric diagnoses"
(Reisner, p. 414). The ruling was considered to be protectionist and
to favor the states police powers over its protective powers.
23


Jones v. United States
Clear and convincing evidence for the deprivation of liberty
was, nonetheless, stronger proof than that afforded the criminally
insane in the District of Columbia. The Addington precedent was
unsuccessfully used by Michael Jones before the U. S. Supreme Court
in his argument against indeterminate confinement as NGRI (1983).
Jones contended that his jail sentence, had he been convicted of
shoplifting, would have expired after one year; he argued that his
hospital confinement should have met the same standard of proof as
Addingtons clear and convincing evidence for civil commitment.
The Court ruled that NGRI acquittees constituted a class of
criminal individuals, with due process requirements different from
civil standards. The 5-4 decision included references to the literature
on psychiatric inability to predict danger by Justice Powell. The
public policy problem of the deprivation of liberty was defined in
dissent by Justice Brennan, joined by Justices Marshall and Blackmun:
The question is whether it is ever constitutional to
sentence persons to psychiatric hospitals indefinitely. The
question is not whether due process prohibits treating
insanity acquittees differently from civil committees...
to be balanced: the governmental interest in isolating and
treating those who may be mentally ill and dangerous; the
difficulty of proving or disproving mental illness and
dangerousness in court; and the massive intrusion on
individual liberty that involuntary psychiatric hospital-
ization entails. (103 SCt. at 3053, 3054)
Libertarian Polstein (1985) concluded that "Jones sounded the death
knell for the constitutional rights of insanity acquittees" (p. 521).
Her critical analyses of the Courts reasoning found little substance
to support the majority ruling.
24


Less analytical in his legal review than Polstein, Singer (1985)
found the Courts reasoning problematic, confusing and distressing.
He doubted the government would ever again have to prove confin-
ability. Margulies (1904) predicted that the Jones decision would
shape procedures for insanity commitment and release for years to
come. Margulies theorized that commitment really has nothing to do
with mental illness or dangerousness, but rather is retaliation:
This truncated review failed to take into account the
powerful punitive urge which the public and its elected
representatives, as well as some of our most noted judges,
have displayed toward insanity acquittees. (1984, p. 793)
Reform in the direction of more restrictive laws followed the
Hinckley (1981) attempt on President Reagans life. Hearings of the
United States Senate were titled "Limiting the insanity defense"
(97th Congress). The United States Supreme Court considered
psychiatric assessment of danger in their 1983 session, reflecting a
public interest in the policy debate.
A provocative notion was formulated by Callahan, Mayer, and
Steadman (1987). Their theory led to the study of legislative
changes made in the three years prior to the U. S. Supreme Court
ruling in Jones and the three years after. The Steadman group
suggested that the Hinckley trial and NGRI verdict of 1982 was not
the moving force behind increased conservatism. They posited,
rather, that Jones legitimized current practices of prolonged and
indefinite hospital stays. Some evidence of legislative response to
the Supreme Court ruling was offered, but the events occurred in
such temporal proximity (Hinckley, Jones, legislation) that their
findings were inconclusive.
25


Less offended by the Jones holding than Singer or Margulies,
Hermann (1983) picked up on a footnote of procedural protection,
that Jones had voluntarily entered an insanity plea to shoplifting.
When the defendant himself raises the defense, automatic commit-
ment may be more justified. Hermann implies the very culpability
the NGRI was relieved from (that is, an absence of mens rea
appropriate to the elements of the offense). Similar to this logic is
the required competency in Colorado on the part of the defendant
before entering a plea of insanity. Hermann supported some of
Justice Brennans dissent, with assurance that states with less
restrictive laws than the District would not be required to conform
with the more restrictive federal ruling about automatic and indeter-
minate confinement.
Schmidt (1984) lamented that Justice Powell's "common sense"
judgments flew in the face of social science research. The majority
wrote "It comports with common sense that someone whose mental
illness was sufficient to lead him to commit a criminal act is likely
to remain ill and in need of treatment" (3050). Schmidt found the
premise untenable in law or in social science. Legal reviews of the
policy implications of Jones tended to be libertarian, and several
suggested that presumption of dangerousness and continued insanity
simply assures confinement.
Legal Definitions of Danger
One policy analyst defined the prediction problem as a broader,
conceptual issue that went beyond law and psychology (Shah, 1978).
In an article suggesting a research paradigm that would distinguish
26


between dangerousness and violence, and between legal definitions
and clinical, Shah reviewed "judiciary" definitions. A federal judge
saw danger in a defendants "check writing proclivities" in Overhol-
ser v^ Russell (1960). In an indecent exposure case, Carras v.
District of Columbia (1962), the court reasoned that sexual danger
was not restricted to physical injury alone. The Carras court wrote
of potential harm including psychological harm to "sensitive adult
women and small children" (1978, p. 226).
A few years later in Millard v^ Harris (1968) the same federal
jurisdiction allowed that only a small minority of "supersensitive
women and small children are likely to suffer serious harm from
isolated instances of exhibitionism" (1978, p. 226) and released
Millard. The U. S. Court of Appeals noted that "a finding of danger-
ousness must be based on a high probability of substantial injury" in
Cross v^ Harris (1969). Judge David Bazelon cautioned against
labeling as dangerous anyone we preferred not to encounter, and
assumed that Congress had not intended dangerous to be used "in any
such Pickwickian sense." (1978, p. 226) Shahs final example was
from the Supreme Court of New Jersey in State v. Krol (1975) and is
relevant to outcome definitions with the Colorado sample.
Dangerous conduct is not identical with criminal conduct.
Dangerous conduct involves not merely violation of social
norms enforced by criminal sanctions, but significant physical
or psychological injury to persons or substantial destruction to
property. Persons are not to be indefinitely incarcerated
because they present a risk of future conduct which is merely
socially undesirable. (1978, p. 226)
Colorados district courts continue to debate whether an alcoholics
self destructive and socially disruptive behavior constitutes an
27


abnormal mental condition likely to cause danger to society. In
Stephens \\ Colorado (1987) the patient claimed that he had feigned
insanity in 1977, and that his alcohol problem did not constitute an
abnormal mental condition that caused him to be dangerous.
The prosecution, in a contested conditional release hearing,
argued that Stephens had been institutionalized (incarcerated for 30
of his 52 years), couldnt fend for himself, and might drink himself
to death or get killed on the railroad tracks in Chicago. A psych-
iatrist testified that the patient was not treatable, did not have a
"disease or defect" and that society could not provide "brothers
keepers" for all such persons. The judge noted that "abnormal mental
condition" included personality disorders as far as he could tell. The
judge acknowledged the clinicians aversion to warehousing people,
but concluded "This statute is primarily designed for the benefit of
protection of the people" (p. 9). The patient was denied release for
his long, disruptive, and essentially nonviolent criminal career.
Outcome Criteria for Danger
First generation research counted arrests, usually making no
distinction among types of criminal charges. Kehospitalization had
been defined by Baxtrom and Dixon researchers as "recidivism". The
Maryland group in 1984 were among the first to report the legal
disposition that followed disruptive or criminal behavior. If one only
counts a subsequent NGRI finding as "recidivism," then extremely low
rates are reported. In Colorado (1977) and for the decade since,
rates of new NGRI commitment were at one and two percent.
28


Conviction and correctional incarceration rates among former
mental patients appear much higher than recommitment as NGRI,
confirming that a history of legal insanity was no guarantee of a
successful plea at a later time to new criminal charges. Mental
health diversion was reported to be common, hut assaultiveness was
the only behavior defined as dangerous. Rubin (1972) defined the
assaultive crimes of homicide, assault, armed robbery, and rape as
violent crimes. Shah (1978) agreed with that as an operational
definition of "violent crimes." Violent behavior on conditional release
is not required, however, for initiation of revocation procedures in
Colorado.
The statutes allow the patient to be declared no longer eligible
to remain on the status for any violation or criminal behavior, as a
result of an abnormal mental condition. That is, the danger criterion
is invoked again in the original court of commitment. Because of the
official nature of revocation, that outcome was defined as a danger-
ous disruption in this research.
One premise underlying the design of this study and the
outcome variables for measuring disruption after conditional release
was an assumption that relapse for a chronically schizophrenic
mental patient is not synonomous with dangerous behavior. When
relapse occurred, in the absence of crimes of violence or revocation,
disruption was classified as nondangerous disruption. Rehospitaliza-
tion through voluntary or civil action was to be expected, and may
have served to prevent further decompensation and dangerous
behavior.
29


Psychological Assessment of Danger
The use of psychological tests for more valid and reliable
measurement of a propensity for danger was described as one
diagnostic approach reported in the literature. Public protectionists
who criticized the early conclusions (overclassification of danger and
the generalizability of findings) were studied for alternative inter-
pretations of the research problems. Floud and Young (1982)
philosophized preventive confinement for those whose dangerous
behavior was "willful harm" and contended that there is no way to
avoid the concept of guilt and just dessert. The British policy study
cautioned, however, against detaining persons for what they might do
in the future on the basis of their thoughts and ideas.
A psychological trait, like aggressiveness or hostility, was
defined as a relatively stable personality characteristic that might be
perceived as dangerousness by the evaluator. Personality traits were
clinically related to the idea of a persons propensity to injure or
harm another. Quite apart from social situations or environmental
circumstances, dangerousness was hypothesized as a personality
characteristic, much like intelligence or aptitude. The concept
implied that some driving, destructive force resides within the
dangerous offender on a permanent basis. Greenland (1985) indicts
psychiatrists for assuming a notion of individual violent tendencies:
Although most physicians recognize the close relationship
between poverty, powerlessness, and disease, the impact of
this association on the practice of medicine has been very
modest. This is especially so in forensic psychiatry,
which still acts as if the impulse to rape, maim, or kill
resides like malignancy within the individual offender.(p.38)
30


Even if such a malignancy could be identified, of greater signif-
icance to prediction accuracy is the presence of similar impulses
among safe individuals. Hostile, aggressive, destructive urges and
ideas may be common to human experience, and only rarely lead to
violent acts.
Personality Scales
Validation studies on trait theory were attempted with per-
sonality tests, and similar psychometric approaches, to try to
distinguish dangerous from safe persons. Megargee (1976, 1981)
postulated that certain items on the MMPI (Minnesota Multiphasic
Personality Inventory) could measure over- and undercontrolled
hostility, each extreme considered as a propensity for dangerous
behavior. A new scale (OH) score was obtained by factor analysis.
The overcontrolled person tended to be passive, prosocial to the
point of rigid values, nonexpressive of emotion, and uncomfortable
with his own and others aggression. Such a person could behave in
violent ways when controls broke down and released pent up
hostility.
The undercontrolled person was thought to be impulsive,
overtly angry, narcissistic, and antisocial in values. His hostility
was not pent up, but was released by his lack of controls. The
underlying construct of hostility was thought to be a proxy measure
for the concept of a trait of dangerousness, with both the over-
controlled and undercontrolled at high risk.
Megargees work was methodologically sound, with replicable
items and scoring techniques. The problem of unacceptably high
31


rates of false positives, that is safe persons who tested dangerous
on the OH scale, was not overcome in subsequent tests of validity
with the MMPI on a number of criminal, mentally ill, and normal
populations. After more than a decade of refinement, Megargees
1970 observation was confirmed. "No methods have been devised
which will adequately postdict, let alone predict, violent behavior"
(1970, p. 145). The finding was based on his inability to distinguish
prison inmates with violent crimes from inmates with nonviolent
crimes, or prison from nonprison samples, some of his tested criterion
variables for danger.
Psychological trait theory had a major influence on attempts to
identify a characteristic of dangerousness or a propensity for
violence within the offender. Clinical tradition in personality
assessment favored a projective technique of the Rorschach inkblots
as a diagnostic test of angry, hostile, and aggressive ideas and
urges. There is little substantive evidence of Rorschach validity,
however, in the law or the literature on prediction of danger
(Webster et al., 1985).
Meloy (1988) a psychodynamic practitioner who works with
offenders in southern California, postulated two types of violent
persons and disagrees with the majority critics of the Rorschach.
Meloy classifies dangerous persons as "affective" or emotional, or as
"predatory", and reports a number of different profiles on standard
psychological tests (1990). He places emphasis on early childhood
bonding experiences that influence motivations of fear and anger
later on in life. Unlike first generation researchers who found
32


diagnosis not very helpful, Meloy and other recent reports have
resurrected Cleckleys (1964) psychopathic profile and associated the
diagnosis with violence. The Bridgewater legacy and recent attempts
at sex offender typology fit a similar philosophical model.
Contemporary studies of violence tend to focus on short-term
classifications for management of acute psychiatric patients in
hospital settings. Yesavage (1984) measured schizophrenic symptoms
on the Brief Psychiatric Rating Scale (BPRS), serum level of a
standard dose of neuroleptic drug, social and psychiatric history, and
disruptions during the first eight days of hospitalization. This
Veterans Adminstration hospital found that Vietnam combat exper-
ience was correlated with in-hospital disruption.
Klassen and OConnor (1988) suggest that we may never achieve
improved accuracy rates for longer range prediction, but they chose a
mid range follow-up of about six months. Theirs is a refreshing
focus on situational variables, albiet not very practical ones. For
example, the amount of time since last intercourse, injuries ex-
perienced at the hands of a sibling before age 15, and the like.
Family, friendship, and work environments were evaluated for
predictive value. With an N of 239 adult males they used discrim-
inant analysis with 67 potential predictor variables. Parent loss
before age 15, self- reported violence, and dissatisfaction with family
were associated with subsequent violence.
The Bridgewater Legacy
One of the first published attempts to evaluate the diagnosis
and treatment of dangerous sexual psychopaths came from the
33


security hospital at Bridgewater, Massachusetts (Kozel, Boucher, and
Garofalo, 1972). Evaluators formulated psychodynamic concepts of
personality and behavior thought to be related to characteristics of
dangerousness. The multidisciplinary team reported solid interrater
reliability, suggesting sophistication in operationalizing their defini-
tions. Similar concensus with a rating instrument (TRIAD) used in a
California psychiatric emergency room (Segal et al., 1986) also
reflected a common philosophical base with good reliability among
raters of short term danger.
Validity of longer term danger assessments, and validity of the
psychiatric nomenclature, is an issue separate from reliability among
raters. Whether either group actually measured dangerousness, rather
than some social value, attitude, biochemical state, or psychiatric
symptom, could be known only through replicable measures. The
interrater reliability suggested common decision rules within each
setting, but did not establish construct validity.
Psychopathy and Validity of Diagnosis
A seminal work by libertarian and activist Bruce Ennis (Ennis &
Litwack, 1974) was a classic commentary on the imprecise nature of
psychology and psychiatry. Ennis showed how nonscientific and
arbitrary psychiatric testimony was. Behavioral criteria for diagnostic
nomenclature improved with the revised, third edition of the Diagnos-
tic and Statistical Manual of the American Psychiatric Association
(DSM III, 1985), but the personality disorders remained an enigma for
practitioners and researchers alike. The diagnosis of psychopath,
sociopath, and antisocial personality disorder, has been referenced in
34


and out of statutory definitions and diagnostic nomenclature as long
as insanity law has been recorded. Clinicians believe it is easy
enough to diagnose, but policy makers disagree on whether it is a
form of mental illness or insanity.
After a career devoted to public sector forensic psychiatry,
Kozel (1982) offered the following definition of the dangerous person,
which sounds much like diagnostic descriptions of psychopathy:
One who has actually inflicted or attempted to inflict
serious physical injury on another person; harbors anger,
hostility and resentment; enjoys witnessing or inflicting
suffering; lacks altruistic and compassionate concern for
others; sees himself as a victim rather than as an aggres-
sor; resents or rejects authority; is primarily concerned
with his own satisfaction and with the relief of his own
discomfort;...has immature attitudes toward social respon-
sibility and distorts his perception of reality in accordance
with his own wishes and needs, (p. 251)
The methodological problem is finding valid measures for identifying
just which offenders, in Kozels words, still "harbor anger" or "see
themselves as victims." The concepts are clinical and interpretive,
rather than behavioral, making operational definition difficult.
The Meloy work, and the British Columbia work (Hart et al.,
1988) with the Hare Psychopathy Checklist, reflect a psychodynamic
philosophy similar to the Kozel group. More recent work from the
same Massachusetts facility (Rosenberg et al., 1988) did report, as
did the Canadian group, more technically sophisticated methodological
approaches. Their predictor variables were quite clearly based on a
similar, nonstatistical classification theory reported by an earlier
Bridgewater group (Cohen, Garofalo, and Boucher, 1971). Impulsivity
and rage are examples of motivating factors that differ among
sexually assaultive persons in their typology.
35


Prediction Accuracy
The Bridgewater population was probably the largest and most
dangerous group (n=435) of psychiatrically treated sex offenders ever
studied (Kozel, 1982). The Massachusetts treatment emphasis was to
help offenders become prosocial and morally responsible. Recommen-
dations were made to the courts about risk level at the time of
release, with recidivism (from arrest records) used as a method of
measuring accuracy of the danger classifications. Offenders predicted
to be dangerous did, in fact, have higher recidivism rates (35 percent
of 49 persons evaluated dangerous) than those predicted to be safe (8
percent of 82 persons).
The Kozel et al. (1972) study is referenced as one of the few
with primary data, and is used by most policy analysts to show over
prediction of dangerousness, with two out of three high risk offen-
ders not displaying dangerous outcomes. Critics complained that the
followup time was too variable; so the Kozel group continued work
with the sample for about 10 years, confirming similar rates among
an accumulated 592 patient evaluations. Gordon (1977) compared the
Massachusetts relatively high (.35) base rate, however, to his study
of Patuxent patients (.41) and a Colorado sample (.48), the proportion
of high risk patients with dangerous outcomes.
Did those samples (Massachusetts, Maryland, and Colorado)
provide enough evidence of external validity to warrant the general-
izations of unacceptably high rates of false positives made in the
secondary literature? Gordon and this researcher think not. Three
reported probabilities were near even chance, a great deal higher
36


than the general population's chances of behaving dangerously.
Considering our clinical ability to recognize safety in a general
population, Gordon contended that to conclude that we cannot predict
danger is also to conclude that we cannot predict safety, when in
fact, "Most of us are predictably quite safe." (Gordon, 1982, p. 296)
Replication of the Kozel groups findings was precluded by lack of
sufficient criteria or decision rules to substantiate their reported
reliability among raters of danger.
The distribution and homogeneity within a sample group of
offenders drives this and any other study lacking cross validation.
The kind of patients at Bridgewater excluded psychotics and included
personality disorders with sexually assaultive crimes. Psychiatric
hospitals are not likely to have such samples for study, at least not
from the present decade. Elimination of the irresistible impulse
from the Colorado insanity test (a volitional test that said the
defendant knew right from wrong, but could not choose the right and
refrain from the wrong) eliminated most defendants with primary
personality disorders (PDS reports, 1981-1989). The present study
included less than a dozen who would fit the description of Bridge-
water subjects.
The Baxstrom and Dixon Cases
A ruling by the United States Supreme Court in Baxstrom Vi
Herold (1966) affected nearly 1,000 mental patients housed in New
York security hospitals. The Court held that inmates who became
mentally ill in prisons had rights to due process at the time of
37


transfer to state hospitals for the criminally insane at Dannemora and
Mattawan. Inmates were entitled to be transferred to civil hospitals
when their sentences expired. Most were manageable in civil
hospitals and many died there. The patients affected were the oldest
in the New York facilities, with almost 40 percent over the age of 70
at the time of the court ruling (Steadman and Cocozza, 1974).
Researchers found only three studies of the criminally insane in
the international literature at the time of the Baxstrom followup in
1970, and one of those was an unpublished study at Colorado State
Hospital (1966). The early studies were the first demographic
descriptions of persons broadly defined as criminally insane. The
mean age of Baxstrom cases was 51.6 years and 45 years for the
Colorado sample. New York cases had an average length of hospital
stay of 14 years, compared to the Colorado report of an average
length of stay of 2.5 years for NGRIs released in 1962-64.
Monahan (1981) called the long term detention of Baxstrom patients a
result of bureaucratic inertia, and Greenland (1985) said the studies
provided "...depressing information about slovenly administrative and
diagnostic procedures." (p. 31)
Criterion validity was based on only 98 (of 969 patients in the
Baxstrom class) who were eventually released to community place-
ments and followed for rearrest or rehospitalization. The 98 Baxst-
rom releases were followed after an average of four years at risk in
communities. Rearrest for index crimes of homicide, rape, aggravated
assault, and armed robbery had been previously used definitions of
Rubin (1972) as both determinants, that is, predictors, and as the
38


outcome or criterion variables. New York researchers learned that
not all assaultive behavior resulted in arrest or criminal charges.
Diversion of former mental patients into mental hospitals was thus
included as dangerous behavior when assaults occurred.
The Legal Dangerousness Scale
The Baxstrom researchers explored dozens of variables kept in
institutional records and New Yorks central Department of Mental
Hygiene, and tested them for predictive association with dangerous
outcomes. Assaultive behavior during hospital career, psychiatric
diagnosis, and hospital length of stay were examples of variables
found not to discriminate the dangerous from the safe. The patients
were old and institutionalized, so history was often remote.
By restrospective statistical analyses of social and psychiatric
events in patient histories, Steadman and the New York group
concluded with a four item Legal Dangerousness Scale (LDS). They
assigned a binary coding score for each item of criminal record,
resulting in a perfectly reproducible summary measure. Reproduc-
ibility came from the fact that only one combination could yield
each score from 0 to 15. Points were summed, and a cutting score
of 5 and age under 50 designated the high risk group (31 percent of
36 patients were subsequently dangerous), with the low risk group
having only 5 percent dangerous outcomes among the 62 patients
predicted to be safe by the LDS and age risk assignment.
The scale was a relatively simple and straightforward approach
that objectively assigned offenders either to a high or low risk
group, according to events of criminal history and age. The New
39


York reports ascribed Guttman type properties of ordinality to the
items, and assigned unequal weights (actually a binary code) to four
categorical events of record. A coefficient of reproducibility of .91
was reported as evidence that the items were scaleable.
The scoring technique that implied that juvenile record (8
points) was twice as important as previous incarceration (4 points),
and constituted automatic assignment to the high risk group, was
questioned in a validation study of the LDS properties by Koppin
(1977). Reported school disruption, previous psychiatric hospitaliza-
tion, and nonviolent crimes were also associated with subsequent
dangerous behavior. The youngest age group had proportionately less
dangerous behavior than men in their thirties and forties.
Ordinality, as least in the Guttman sense, was not confirmed in
the validation study with an N of 111 and a coefficient of reproduc-
ibility of .83. The published Baxstrom reports, including the Cocozza
dissertation of 1974, did not describe the research method used for
derivation of weights, and unequal weights were not confirmed in the
replication work. The scoring system did have value simply as a code
for identifying which variables were present in the offenders history.
There was only one possible combination of LDS items for each score:
Juvenile record = 8
Previous incarceration = 4
Prior violent crime = 2
Severe instant offense = 1
Guttman properties suggested that if the first item were present, all
remaining items would also be present. That was not confirmed, but
40


predictive value of the IDS variables was found with the Colorado
sample. The LDS high risk group of 60 had 48 percent dangerous
behavior after release, with only 8 percent of 51 low risk patients
being dangerous. Base rates of total samples were remarkably similar,
but various combinations of criminal and social histories yielded about
the same predictive value as the New York scale items. At best,
prediction of danger was accurate in about half of the cases of high
risk, and total accuracy rates hovered near the base rate of the
criterion behavior.
There were few empirical investigations, and even fewer
techniques for comparing samples, methods, or findings, when the
Legal Dangerousness Scale was published in 1974. Policy analyst
Gordon used the scale as a comparative statistic in public policy
studies of Marylands defective delinquents in 1977 and 1982.
Correctional literature has continued to use criminal history as one
of the best discriminators for identifying parole success and criminal
recidivism.
Political Predictions and Dixon Cases
Strikingly similar to Baxstrom findings, mass transfers of
patients from Pennsylvania security hospitals to civil settings followed
a court decision in Dixon v^ Attorney General of the Commonwealth
of Pennsylvania (1971). These 586 patients were also not specifically
predicted to be dangerous, although 114 were later opposed for
transfer on those grounds. All Dixon patients had been held in
secure hospitals for an average of 14 years, without appropriate due
process under the Fourteenth Amendment.
41


Researchers broadly defined prison transfers as criminal commit-
ments, along with those under a defective delinquent law for indeter-
minate sentences. The affected Dixon class had become civil
recommitments after expiration of criminal statuses, without equal
protection. Fourteen percent (of 438 patients) were considered
dangerous after a four year follow up by Thornberry and Jacoby
(1979). Like Baxstrom cases the mean age was 47 at the time of
transfer and many had died by the time of follow-up.
Unique in the literature, the Pennsylvania researchers followed
their patients and asked them about their hospital experiences,
current living situations, employment, finances, and general well
being in community placements. A replicable aspect was their use of
the Katz Adjustment Scale (Katz and Lyerly, 1963) for clinical
adjustment and level of functioning measures. The instrument had
also been normed on nonpsychiatric populations, and had baseline
data on "typical" mental patients. Dixon patients did not score well
on these indices, but were uniformly happier being out of the
security hospital than in.
Prediction Accuracy
Thornberry and Jacoby (1979) retrospectively applied the New
York Legal Dangerousness Score, with less discrimination found for
identifying the dangerous among the Dixon class. Several years
passed before the results were published amid some skepticism
surrounding other right-to-treatment cases in Pennsylvania. Their
study is the only one known to have included "self-releases" by
escape (p. 193). Twenty-four percent of all releases were arrested at
42


least once, and 14 percent were identified as dangerous, defined as
behavior injurious to another person. Time at risk was as little as
less than one month to over four and a half years.
Pennsylvania researchers also used assaultive behavior in the
civil hospitals as an outcome criterion, and found that the security
hospitals assessments of danger were exaggerated. The authors
concluded that the decision making processes were not actually
clinical, nor statistical, but rather political and concerned with
political consequences:
We would submit, however, that there is a third type of
predictive exercise...which we will call political prediction.
It is based, not on the characteristics of the individual,
but on the assumed characteristics of a group to which the
individual belongs. In essence, group predictions are
projected onto the members. When the process is properly
conducted the probability of group events, and the deduced
individual event, is based on grouping subjects together on
the basis of variables known to be associated with offend-
ing. In political prediction, groups are created on the
basis of criteria which are assumed, but not demonstrated,
to be correlated with offending (Thornberry & Jacoby,
1979, p. 27).
Examples of such groups are the mentally ill, felons, murderers or
rapists, psychopaths, schizophrenics, or perhaps in the present study,
NGRIs.
Research Propositions
Few studies with precisely defined samples according to NGRI
legal status were reviewed here. Specific statements about a
patients likelihood to be dangerous, as a result of an abnormal
mental condition, were not operationalized in the clinical or legal
literature. Clinical predictions were either inferred retrospectively
43


or were vaguely related to unspecified and nonreplicable treatment
programs. Nearly all were based on zero-one probability models of
dichotomous prediction and outcome groups.
Assignment to risk groups on the basis of statistical frequencies
typically yielded better rates of predictive accuracy than clinical
methods. It was assumed that methodological approaches to evaluat-
ing accuracy of danger classifications were as relevant to rational
policy debate as predictor items. The following research statements
were formulated to test methods and statistical techniques, as well as
to test for variables to assign patients to high or low risk groups.
Models of actual and equal probability, and measures of improvement
over random distribution (chance) were the statistical approaches to
the research questions identified in the review of the literature.
Proposition 1. Agency opposition to release constitutes a
prediction of danger and will distinguish the dangerous
from the safe after release.
Proposition 2. Groups defined by agency recommendation
for or against release will themselves be distinguishable by
disruptive events in patient histories.
Proposition 3. Disruptive events in precommitment
histories and hospital careers will distinguish the dangerous
from the safe after release.
Age, sex, race, criminal record, and certain in-hospital events were
tested for predictive value in identifying the dangerous after release.
Statistical models are described in the next chapter.
44


CHAPTER 3
METHODS
The study was a retrospective analysis of the outcomes of 109
NGRI patients who obtained conditional releases from the forensic
unit of Colorado State Hospital between 1981 and 1986. Hospital
recommendations to the criminal courts of commitment constituted a
clinical prediction of each patients likelihood of future danger, as a
result of an abnormal mental condition. NGRIs who were released by
the courts against hospital recommendation (n=31) were hypothesized
to be at higher risk for danger than NGRIs whose release was
recommended (n=78). The release cohort was followed after a
minimum of two years at risk in the community, for psychiatric
relapse and criminal recidivism.
Research groups were thus defined by clinical prediction and by
type of outcome. Revocation of conditional release and/or subsequent
crimes of violence were defined as dangerous disruption after release.
Nondangerous disruption included rehospitalization or rearrest for
nonassaultive behaviors or relapse. The design, including the predic-
tion models replicated from previous research, examined some
theoretical and empirical relationships among pre-hospital, in-hospital,
and post-hospital disruptions. Statistical relationships among disrup-
tive histories and release outcomes were explored to find methods to
discriminate dangerous from safe patients. Details of the research
process are described under the following headings:


1. Design and Prediction Models
2. Selection of the Sample
3. The Treatment Setting
4. Description of the Sample
5. Release Procedures
6. Data Sources and Limitations
7. Operational Definitions
8. Statistics
Design and Prediction Models
Like the Steadman-Morrissey (1981) study of New York incom-
petent to proceed cases, the data analyses were retrospective, but
predictor variables were applied as if they constituted prospective
scales for high and low risk. The Steadman-Morrissey approach tests
the proposition that actuarial approaches would have been as accurate
as clinical predictions, and that either would have been more accurate
than chance. Variable reduction was achieved with discriminant
function analyses of predictor items to distinguish dangerous from
safe outcomes as a dependent variable. Similar analyses of the items
were applied to the dependent variable of hospital opposition (that is,
the prediction of danger). Steadman and Morrissey (1981) compared
their design to a traditional cross validational model.
A Bayesian (actual) probability model consistent with a civil
libertarian perspective, and an equal probability model consistent with
a public protectionist perspective, were applied to determine the
better accuracy and the social costs of one model over the other.
46


The libertarian model emphasized the goal of reducing false positives;
the protectionist model emphasized the minimization of false
negatives.
Predictions, or assignment of high or low risk for future
danger, were treated as independent variables. The dependent
variable, dangerous behavior after release, was deduced from disrup-
tive outcomes, including civil and voluntary rehospitalizations. A
two-by-two contingency table shows the prediction (independent)
variable in the columns and the outcome (dependent) variable in the
rows.
Risk Assignment Groups
Prediction
Safe Dangerous
Outcome i
True I False
Safe Negative Positive
A ! B
False ! True
Dangerous Negative i Positive
C I D
Figure 1. Prediction by Outcome Contingency Table.
A. True Negatives: persons (accurately) predicted to be
safe who were, in fact, safe after release.
B. False Positives: persons (inaccurately) predicted to be
dangerous, who, in fact, were safe after release.
47


C. False Negatives: persons (inaccurately) predicted to be
safe who were, in fact, dangerous after release.
D. True Positives: persons (accurately) predicted to be
dangerous who were, in fact, dangerous after release.
Additional analyses involved a similar, but two-by-three,
contingency table with three outcome groups, in order to capture
disruptions in the community that were not associated with violent
crime or revocation. The dependent variable in those tables thus
became three rows for NO DISRUPTION, NONDANGEROUS DISRUPTION, and
DANGEROUS DISRUPTION.
Selection of the Sample
A group of 144 NGRIs who obtained their first release from
criminal court commitment after June 30, 1981 and before July 1,
1986, were identified through the Patient Data System (PDS) micro-
computer search. Nineteen of 24 unconditional discharges were
NGRIs released to law enforcement detainers to face trial or serve
sentences on prior criminal convictions. Direct discharges conse-
quently had less opportunity for criminal recidivism in the community
than conditional releases. The researcher also had less information
on which to evaluate outcome. Discharge by death (8) and uncond-
itional discharge (24) were thus excluded from the follow up sample,
as were (3) conditional releases lost to follow-up.
A sample of 109 males conditionally released for the first time
from NGRI commitment was selected. Included were patients with
sporadic contact with the hospital, mental health centers, families,
48


police, or fellow patients. The majority remained on conditional
release status for at least one year, with the aftercare staff provid-
ing written progress reports to their courts of NGRI commitment.
Outcomes from all sources were included in the data when validated
by law enforcement or hospital records.
A small group of continuously enrolled re-releases, recycled on
former court orders of release, were excluded as atypical of the
release process. Wide variation in their lengths of initial stay, and
subsequent rehospitalizations ranging from days to years, made
calculating time at risk awkward and cumbersome. Also excluded
were two cases whose conditions of release included "signing in
voluntarily" for continued forensic hospitalization. NGRI women
were excluded from the sample because they were in nonforensic
programs.
All remaining conditional releases were identified from PDS
listings of status and enrollment changes between June 30, 1980, and
July 1, 1986. The final sample included 109 first conditional releases
from NGRI commitments, placed in communities rather than other
correctional institutions, with opportunity for psychiatric relapse or
criminal recidivism. Diagnostic and demographic characteristics are
described in Description of the Sample.
The Treatment Setting
The Colorado State Hospital is fully accredited by the Joint
Commission on Accreditation of Healthcare Organizations, a private
professional standards agency of the medical professions. Hospitals
49


voluntarily subscribe for rigorous surveys and evaluation for com-
pliance with JCAHO standards. The hospital is state licensed as a
700 bed public mental hospital in Pueblo, Colorado. At the time of
study, the hospital was organized into clinical divisions, the largest
one a 350 bed specialty unit for forensic psychiatry. This division
was named the Institute for Forensic Psychiatry (IFP) and, at the
time of the inpatient careers of this NGRI sample, was organized
around three specialty programs located in five buildings.
Treatment Tracks I and II had physical facilities at each of
four security levels. High security levels were maximum and medium
units with fences, electronic surveillance, and guards. Low security
levels were intermediate and minimum units, similar to nonforensic
adult psychiatric wards. Ward level care was given by registered
nurses, mental health workers, and licensed psychiatric technicians.
Female staff were involved in clinical treatment on all security levels.
The male to female staff ratio was about 70/30 during the study
period.
Program Description
Track I was the larger treatment program with an average
daily NGRI census of 120, compared to 90 NGRIs on Track II. Each
program also treated incompetent to proceed to trial and civil
patients, and Track II had several convicted sex offenders under
parole board transfer. In any typical daily census, NGRIs comprised
about two-thirds of the total number of IFP patients. The Track I
program was for chronically mentally ill persons, often psychotic and
socially inadequate, or debilitated by symptoms of institutionalization.
50


Supportive, individualized programs for daily living and basic com-
munication were formulated by core professional staff for continuity
of care along security levels. Antipsychotic drugs were commonly
prescribed to bring patients into sufficient remission for home groups,
problem solving, and reality therapy. Didactic material on major
mental disorders was routinely included in small group therapies.
Psychotropic medication was less routinely used on Track II
because of the diagnostic case mix, although about a third of the
Track II patients were well compensated from acute psychoses. While
many Track II patients had long, antisocial histories, they were more
socially outgoing and verbal than those classified as Track I. Track
II patients often carried secondary diagnoses of personality disorders
or substance abuse. A therapeutic community model of patient
government, mutual responsibility, peer confrontation, and insight
oriented psychotherapy was followed across Track II security levels.
Self help groups, didactic programs, and psychological literature were
available to Track II patients from medium security forward.
As patients developed their own internal controls and complied
with treatment expectations, they were rewarded with increased
privileges and transferred to lower levels of external security.
Gradual increments in the privilege process were common to both
treatment programs, with incentives for cooperation in treatment.
Both programs transferred patients to higher levels of security when
believed necessary for external control and management of illness or
behavior. Behavior management transfers (BMTs) were defined as
in-hospital disruptions, along with escapes, and leaves for jail or
51


court appearances. Track I patients spent proportionately more time
on low security areas and Track II patients spent proportionately
more time on high security areas.
Prerelease Testing in the Community
Permission from the court of commitment was required for any
treatment or rehabilitation activities off of the hospital grounds.
District attorneys had an opportunity to contest the granting of a
court order sanctioning such activities, and arguments were subse-
quently heard by the judge. Progression to unsupervised passes in
the community was implied by clinical and administrative reviewers as
a necessary release criterion for patient release eligibility. Passes
were used for work, school, home visits, or approved recreation in
the local community.
Passes were generally considered to be a time for gradual re-
entry and testing of the patients internal controls. Patients
were expected to bring back to therapy the experiences and accom-
plishments from their time away from the structure of the hospital.
Abstinence from street drugs and alcohol, and discretion in sexual
relationships were encouraged for patients in both programs. Day
passes were given in the beginning, followed by overnight, weekend,
and a week or more. Supportive or independent living arrangements
were used in the community, according to the patient's level of
functioning.
52


Maximum Security Programs
The third program, sometimes called Track III but located in
only one physical facility, was the program of Special Services, a
maximum security unit for all IFP admissions. Acute care and short-
term evaluations were housed on a four-ward, 80-bed security
hospital built in 1974. A yearly intake of some 600 patients included
civil cases from county jails, criminal court pretrial evaluations,
correctional transfers, incompetent to proceed to trial cases, and civil
commitments considered unmanageable by the civil hospitals.
About a dozen highly dangerous or disruptive NGRIs were in
maximum security treatment programs, and a few of those were
discharged to correctional detainers during the time period studied
here. They all had been tracked and treated in less secure program
assignments, and had then been transferred (by BMT) to maximum
security prior to release.
Description of the Sample
Forty-four NGRIs were conditionally released from treatment
Track I, with 7 official agency oppositions to their release. Sixty-
five patients were released from treatment Track II, with 24
releases opposed by the agency but approved for release by the
courts. Most patients (64 percent) were released from low security
treatment areas, with the hospital opposition rate higher for those
seeking release from high security areas.
The sample is representative of only a small proportion of
NGRIs treated during the years of study. In one sense, the sample is
53


a "universe" of all first conditional releases within a certain time
frame. Their commitment origins spanned many decades, however.
The better method for sampling typical practices would have been to
take an admission cohort. Greater generalizability could have been
achieved with patients admitted under similar policies and programs,
rather than release cohorts like these samples.
Psychiatric diagnoses given by psychiatrists at the time of
examination for release confirmed significant program differences.
Primary psychoses were diagnosed for 84 percent of Track I and 43
percent of Track II patients. Substance abuse was primary for less
than 5 percent of Track I, compared to 29 percent of Track II
patients. Other nonpsychotic disorders were more prevalent on II,
and in general, patients with personality disorders were more often
opposed by the hospital than those with major psychotic illnesses.
Demographic Characteristics
The Denver metropolitan counties accounted for 56 of 109
commitments. Age at the time of release ranged from 19 to 64 years,
with an average age of 34. Ethnic minorities included 20 percent
Hispanic and 15 percent Black patients. Crimes of violence (homicide,
sexual assault, aggravated assault, and armed robbery) were involved
in 61 percent of the NGRI commitments, with Track II having
proportionately more crimes of violence (71 percent) than Track I (48
percent). Nearly half of the total sample had never been married, a
high risk indicator reported by Klassen and OConner (1988). Poor
job histories were reported for patients from both treatment tracks.
Track II patients had more high school graduates and persons who
54


had attended college (a cumulative 60 percent). Their level of social
functioning was one rationale for the cognitive program approach.
Length of Hospital Stay and Crime
The average length of stay was 5.4 years, ranging from the
shortest of 3 months to the longest of 17 years. The extremes in
length of stay were infrequent, with the median at 55 months, or
just under 5 years. Previous NGRI release studies from this hospital
reported 2.5 years in 1966 and 3.7 years in 1977.
Limitation on any generalization is reflected by the small sample
size, and the frequencies appearing in the categories used to describe
the most serious crime of commitment. Average lengths of stay by
crime category appear in Table 3.1. Lengths of stay less than one
year were more common in the early years when there were 12 short
stays.
Table 3.1. Crime of NGRI Commitment by Length of Hospital Stay
in Months: Conditional Releases 1980-1986. (n=109)
Crime n Min. Max. Mean S.D.
Homicide 12 23 199 95.33 53.06
Sexual Assault 10 34 184 81.20 47.79
Aggravated Assault 32 10 185 69.00 43.37
Kidnapping 2 20 36 28.00 11.31
Armed Robbery 10 16 178 66.40 53.17
Simple Assault 8 5 76 39.38 25.01
Arson 5 3 122 39.60 47.69
Criminal Mischief 2 9 59 34.00 35.36
Burglary 14 7 134 60.57 41.82
Trespass/Theft 11 5 203 54.18 54.23
Forgery/Fraud 2 26 41 34.50 9.19
Nonassaultive Sex 1 83.00
Total 109 3 203 64.78 46.49
55


A total of 44 patients accounted for 100 escapes, usually
walkaways from privileges. The average length of time on escape
was 247 days, skewed by a few escapes of several years duration.
Length of hospital stay included time absent on escape, pass, or
leaves to court. In summary, length of stay was not a straightfor-
ward measure.
Release Procedures
The interface between legal and clinical mandates to protect
public safety and provide psychiatric treatment and rehabilitation,
requires complex tradeoffs between social costs and benefits.
Decisions to detain some NGRIs and to conditionally release others
was considered a risky business by clinicians and courts. Implied
tradeoffs were retrospectively evaluated for this study from liber-
tarian and protectionist perspectives, after actual decisions were
made. The study of legal and clinical criteria that may have been
used by decision makers helped define the research problem. One
consideration for psychiatrists and lawyers was the difference
between release criteria and commitment criteria.
The Legal Release Criteria
The proof of insanity at the time of the initial trial did not
require the proof of dangerousness, nor proof of the necessity of
hospitalization. Although NGRI commitment law required no presump-
tion of future danger for initial aquittal or automatic hospitalization,
once adjudicated in Colorado, the law assumed an abnormal mental
condition caused the NGRI patient to be dangerous and in need of
56


hospital confinement. Temporary insanity was not recognized in the
Colorado statutes, and at the time of study a minimum of six months
in the hospital was required before the patient could file a motion
for a release examination and hearing.
In addition to the presumption of a continuing abnormal mental
condition, Colorado NGRIs were legally presumed to be dangerous
because of the release eligibility law, not the insanity commitment
law. A broad definition of future dangerousness is reflected in the
release statute 16-104-18 that requires that the patient no longer
suffers from an abnormal mental condition "...likely to cause him to
be dangerous to himself, to others, or to the community if at large."
In another part of the statutes, the older or more archaic term
of "mental disease or defect" is used. Legal consultants suggested
that legislative intent had to do with the cognitive test of defect of
reasoning. Psychologists have suggested that the mental disease or
defect phrase is more restrictive to major mental disorders than the
broader term of abnormal mental condition. The researcher here used
the one most cited by district court judges in trial transcripts and
court orders for examination, the abnormal mental condition.
Patient Petition for Release
The hospital was under no obligation to review for release eligi-
bility unless the patient filed a motion with the court of commitment.
The sample of 109 conditional releases represents approximately 10
percent of all hospital release examinations for criminal court review
during the time period studied. Most examinations were the result of
the patients legal motion, rather than by team recommendation. In
57


one conservative year, for example, there were 95 patients reviewed
with only 11 having been initiated by the clinical teams. Track II
accounted for more self-initiated release hearings, probably because
of their better social competency.
Psychiatric Recommendations
Primary clinical criteria for release eligibility at the Colorado
hospital was the expectation that the patient discuss his crime of
commitment, understand his legal status and diagnosis, express some
remorse for his violent acts, recognize his own danger signals, and
plan alternate methods of expression and problem solving. All of
these abilities required some minimal level of healthy, prosocial,
interpersonal communication and understanding. Therapeutic relation-
ships with staff and peers were, therefore, crucial to the release
process.
Prior to administrative and court review, a psychiatrist from
within the agency provided a written, formal examination of the
patients mental status and prognosis for future danger. Patients
with lifelong mental disabilities requiring the use of antipsychotic
medication, were expected to be candid about their symptoms (for
example, the frequency and content of hallucinations) and their
response to medication (preferences, side effects, improvements).
The treatment goal was to teach the patient how to manage his own
illness or impulses, with fewer and fewer external controls.
The Administrative Review
Patient petitions and team recommendations were all reviewed
by the hospital's multidisciplinary review board, the Disposition
58


Committee.
Geriatric NGRIs and female NGRIs housed on the
general adult civil unit of the hospital also were under the juris-
diction of the Disposition Committee for administrative and clinical
review. After the Committee privately reviewed the written history
and psychiatric recommendations, the patient was usually interviewed
by the group. Rarely were attorneys present, but the patients
primary therapist usually attended to answer questions and support
the patient.
The board exercised considerable influence within the organ-
ization and the courts of commitment, and its legacy dates back to
the early 1960s. The board members then included law enforcement
officers and politically appointed prominent citizens, whereas today
only hospital clinicians sit on the board. As in the beginning, the
Committees official mission is protectionist with a goal of minimiz-
ing false negatives, that is, inaccurate predictions of safety.
The psychiatric opinion became part of the written Committee
packet sent to the superintendent, the hospital's chief executive
officer and herself a psychiatrist. The boards evaluation was
written by the Committee chairman, a clinical psychologist, who
described the patients psychiatric history, criminal record, course in
hospital, and behavior in interview.
The Official Agency Recommendation
The agency concluded with one official position in the form of a
cover letter by the chief executive officer and superintendent, also a
psychiatrist. Her recommendation constituted a prediction. The
letter to the NGRI court of commitment gave identifying information
59


and the date of release examination, followed by a standardized
opinion:
After reviewing the defendants record and attached
reports, it is my opinion that this defendant continues to
suffer from a mental disease or defect which is likely to
cause him to be dangerous to himself, to others, or to the
community in the reasonably foreseeable future (CRS 16-8-
115, as amended). Therefore, it is the undersigned's
opinion that the defendant is not eligible for release from
hospital care and treatment.
This example is of an opposition letter, estimated to have been the
official agency stance on about 85 percent of all release hearings.
When the agency recommended release the courts nearly always
concurred. There were only 31 NGRIs conditionally released over the
hospitals opposition in the six year study period.
Adversarial Release Hearings
The district attorney in the jurisdiction of the commiting court
could contest any release, whether recommended or opposed by the
hospital. The D.A.s practice was to request the court to order a
second psychiatric opinion, occasionally a doctor of the prosecutors
choice. The process slowed down and sometimes aborted the release
process. When the patient hired an outside psychiatrist the tes-
timony was provided directly to the court as independent of the
Committee and hospital. Courts generally attended to written reports
when no one was recommending release, and heard testimony from
expert witnesses when a recommended release was contested.
Colorado law defined psychiatric technicians, nurses, teachers,
and mental health workers as experts, in addition to psychiatrists and
psychologists. It was not uncommon for the officials representing the
60


hospital position to have members of its own staff on the other side
of the courtroom. Jury trials were infrequent, but were more likely
when there were conflicting opinions and when the district attorney
was opposed to release. The burden of proof was on the patient to
convince the court that he was no longer insane or dangerous.
Data Sources and Limitations
Social and psychiatric histories were abstracted and coded from
medical and legal records of the hospital. Preadmission arrest
records were obtained by the agencys police department from the
Federal Bureau of Investigation (FBI rap sheets) with fingerprints
taken on admission. Similar arrest records were obtained for
follow-up in September, 1987. Crimes committed during the hospital-
ization, while in custody or on escape, were usually entered in
subsequent NCIC records by the arresting authority, and required
close attention to coding dates. Reliability checks were made by
examining the dates of patient absences and locations.
Presence of in-hospital criminal charges was discovered late in
the study, but the variable may well have been a good identifier of
those NGRIs who were later dangerous. In the small number of
cases where no arrest record existed, even though there was a
known instant offense and court commitment, the researcher relied
on the medical and court records kept by the hospital and mental
health centers for aftercare.
61


Patient Data System
Items of criminal and psychiatric history reported in previous
research as associated with the assessment and prediction of danger-
ous behavior were included in the 1980 IFP Patient Data System
(PDS). Individual patient records from PDS were used to create a
research file for microcomputer use and analyses by SPSS/PC+
software. Selected subjects and variables were transferred to the
Universitys mainframe for access to SPSS-X software. Names and
identifying numbers were eliminated for subsequent aggregate
analyses.
Hospital Records
Outcome data came from follow-up variables kept in individual
aftercare records, court reports, arrest records, and from obser-
vations of clinicians, patient families, and other patients. An
aftercare coordinating nurse, William Leimeister, was interviewed at
several phases of the project for clarification of anecdotal notes and
formal reports to the courts. Data were abstracted, coded, and
incorporated in the project microcomputer and mainframe files.
One major limitation to any study on the prediction of danger-
ous behavior is the fact that those who are clinically defined as
dangerous do not usually get released with opportunity for recidivism.
The present sample only captured 31 in six years who were released
over hospital opposition. The methodological problem was that there
was no similar outcome criterion on which to evaluate the accuracy
of predictions on patients who remained under the external controls
of the hospital. Those who did not get out of the hospital, with or
62


without hospital recommendation, limit the generalizations to be made
from release samples.
Level of Measurement
Most data in this research were at a categorical level of
measurement. There were few continuous measures, and no instru-
ments with items of statistically derived weights, available in the
literature for direct measurement of danger. The lack of continuous
level of measurement not only restricted the prospective identification
of high risk offenders, but restricted comparative research on
dangerous outcomes.
Experts repeatedly report that actuarial and statistical methods
of prediction are better than clinical; yet what rudimentary instru-
ments there are contain clinical and interpretive items such as
"manipulative, affective, or predatory" (Meloy, 1989). The weakest
factor in building and contributing to a knowledge base is the
problem of operational definitions for comparative study and repli-
cation.
Validity and Reliability
Contrary to researcher expectations, arrest records consistently
under reported criminal arrests known to the hospital aftercare nurse.
Valid, reliable, and replicable outcome variables to define psychiatric
relapse and criminal recidivism, and to distinguish disruptions related
to abnormal mental condition, were elusive. Much trial and error, as
well as case-by-case judgment, were required before settling on a
definition of dangerous behavior after release. Concern with under-
63


reporting was reduced by including all other possible data sources as
worthy of verification efforts.
Severity of outcome disruptions became a problem for class-
ification of behavior as safe or dangerous. Arrest or rehospital-
ization for crimes of violence was a straightforward definition of
dangerous behavior. Other disruptions resulted in arrest or rehos-
pitalization without revocation of conditional release, and some
involved potentially violent acts like property destruction or menac-
ing.
A number of cross tabulations and experimental combinations of
disruptive events and legal dispositions (discussed further in Chapter
4) led to the inclusion of all revocations in the definition of danger-
ous behavior after release, even though the criminal or psychiatric
disruptions were not always assaultive or violent. Arrests for index
crimes of violence, whatever the disposition, constituted danger
because of the severity of the initial charge of arresting officers.
Revocation required psychiatric testimony on the release eligibility
criterion, thereby legally classifying the patient as again dangerous.
Operational Definitions
Outcome variables were designed to describe behavioral events
after release in terms that could be replicated on other samples of
NGRI patient offenders. Events were further classifed according to
the type of external controls used in response to NGRI behaviors.
Events of criminal recidivism and psychiatric relapse, and consequent
legal dispositions by arrest or rehospitalization, were defined.
64


Disruption After Release
Arrest. Overnight or longer detention by police, including
taking the patient into custody for forced psychiatric examination by
warrant of arrest from the court.
Rehospitalization. Overnight or longer admission to any
psychiatric hospital, under civil or criminal statutes.
New Crime. Category of criminal offense reflecting charges
filed at (re)arrest. Violent= homicide, sexual assault, aggravated
assault, armed robbery. Nonviolent= menacing, arson, criminal
mischief, burglary, trespass, forgery/fraud, disorderly, traffic, status
offenses.
Assault. Crimes of violence and assaults on another person,
with or without criminal charges filed, involving arrest or rehos-
pitalization.
Threats. Menacing with weapons; stated intent to harm, kill,
bomb, burn, etc., involving arrest or rehospitalization.
Relapse. Psychotic decompensation, disabling symptoms,
involving arrest or rehospitalization.
Off Meds. Noncompliance with prescribed neuroleptics, generally
diagnosed psychotic and chronically mentally ill, involving arrest or
rehospitalization.
Abuse. Use of alcohol or other nonprescription drug, with
resulting disruption in interpersonal, employment, or illegal acts,
involving arrest or rehospitalization. Generally in diagnosed sub-
stance abusers.
65


Disposition After Disruption
Revocation. Court ordered termination of conditional release,
on the ruling that patient is no longer eligible (not dangerous as a
result of an abnormal mental condition) to remain on release. The
patient is returned to his former NGRI legal status.
NGRI. New criminal court commitment on new finding of
insanity to criminal charges acquired after release.
ITP. New criminal court commitment as incompetent to
proceed to criminal charges or proceedings initiated after release.
MHC. Mental health commitment under civil 27-10 procedures.
Includes voluntary rehospitalizations.
Corrections. Any disposition following new criminal charges
that involves criminal probation, incarceration, community correct-
ions, deferred prosecution or sentence. Assumes patient account-
ability and conviction.
Prehospital Disruptions
Juvenile History. Arrested for criminal behavior before the
age of 18, with some evidence of juvenile adjudication, probation,
or detention for serious problems with the law Excludes status
offenses or acts not considered criminal if commited by an adult
(runaway, truancy).
Prior Violent Crime. Officially charged and on arrest record,
whether convicted with correctional disposition or a mental health
diversion. Crimes of violence = homicide,sexual assault, aggravated
assault, armed robbery.
66


Prior Incarceration. Any correctional sentence to jail, prison,
or reformatory over 60 days in lock-up. Excludes suspended sen-
tences and jail time pending trial on instant offense, and sentences
served by probation.
Violence of Instant Offense. (Rank order is approximate
severity on face validity only, most to least serious.)
Homicide
Sexual Assault
Aggravated Assault
Kidnap
Armed Robbery
Menacing, simple assault
Arson
Criminal Mischief
Burglary
Trespass/Theft
Forgery/Fraud
Nonassaultive Sexual
Status Offense
Disturbance/Disorderly
Resisting Arrest
Traffic Arrests
Use of Weapon in Instant Offense. In police reports of crime
leading to NGRI commitment, use of a knife, gun, other tools for
attack or intimidation.
Use of Alcohol in the Instant Offense. In police reports or by
self-report, the NGRI was under the visible influence of alcohol at
the time of the instant offense.
Use of Drugs in the Instant Offense. In police reports or by
self report, the NGRI was under the influence of nonprescription
mood altering chemicals, other than or in addition to, alcohol.
Prior Treatment History. History of previous psychiatric
inpatient treatment; history of alcohol abuse; history of drug abuse.
67


In-Hospital Disruptions
Stay. Months in hospital from date of admission to date of
exit on conditional release.
Time. Months in NGRI status only; excluded incompetent time
or civil time under definition of "stay" above.
Maximum. Number of days accumulated on highest level of
security.
Medium. Number of days accumulated on medium security.
Intermediate. Number of days accumulated on (locked) wards.
Minimum. Number of days accumulated on IFP (open) wards.
Special Leave. Number of absences to court or corrections.
Days Authorized Leave. Number of days accumulated on special
leave to custody of court or correctional agencys.
Passes. Number of times placed on overnight or longer pass
status.
Days on Pass. Number of accumulated days on pass.
BMT. Number of Behavioral Management Transfers to higher
security level.
N Escape. Number of times absent without authority.
Escape Days. Number of accumulated days on status.
Treatment Track. Wards 5, 7, 9, 11, 79 = Track I. Wards 2,
10, 12, GW 5,11 = Track II. Temporary locations recoded.
Diagnostic Categories. Psychotic diagnoses = schizophrenia,
organic psychoses, bipolar disorders, atypical psychoses, paranoid
states. Substance diagnoses = substance induced organic, alcohol
abuse or dependence, drug abuse or dependence. Other nonpsychotic
68


= personality disorders, anxiety neuroses, organic personality syn-
drome, mental retardation, adjustment reactions, sexual deviation.
Coded first in DSM III nomenclature to five digits, then categorized.
Interval Level (above) Variables. Subsequent collapse of time
intervals for absences, length of stay, time at risk, into logical
intervals. Median used for some dichotomies.
Legal Dangerousness Scale. New York Department of Mental
Hygiene, no copyright, 1974 Cocozza & Steadman; Juvenile record = 8
points; Previous incarceration = 4 points; Prior violent crime = 2
points; Crimes 1 to 11 serious instant offense = 1 point.
Statistics
The hypothesis that clinical assessment and prediction would be
more accurate than chance distribution of risk was first tested by
nonparametric methods of dichotomous predictions and dangerous
outcomes. Independent variables (predictor items from patient
histories) were taken from findings reported to be associated with
dangerous behavior. Interpretations of types of errors were made
according to the policy perspectives of the civil libertarian and public
protectionist models.
Distributions were analyzed for total accuracy, as well as by
Relative Improvement Over Chance (RIOC) to compare accuracy and
type or direction of errors (i.e. inaccurate predictions of safety or
inaccurate classification as dangerous). Selection factors were
analyzed for optimal cutting scores for classification of high and low
risk. The selection percent in RIOC computations is the proportion
69


of the sample captured by the item or score under consideration.
The formula for RIOC is dependent upon the known frequencies in
the distribution.
RELATIVE IMPROVEMENT = ACHIEVED ACCURATE RANDOM
OVER CHANCE MAX POSSIBLE RANDOM
The maximum possible total correct predictions, in the two-by-
two contingency tables used here, is constrained by the actual row
and column totals. Total achieved accurate predictions are the true
positives plus true negatives. Percent total accuracy is the sum of
accurate predictions divided by N (109) times 100. Maximum possible
accurate predictions are constrained by actual outcomes, in this case
row totals. That is, the calculated theoretical "max possible" can
only include the minimum number of patients who were actually
dangerous after release.
Bayesian probabilities are a form of statistical inference based
on a theorum of equality among prior events, proposed by an English
clergyman in 1763, Thomas Bayes (Iversen, 1984). In the SPSS-X
discriminant analysis, the prior probabilities were set equal to the
proportional distribution, that is, .80 = SAFE and .20 = DANGEROUS.
The resulting functions were then compared to an equal probability
model with a .50 SAFE and .50 DANGEROUS proportional distribution.
The research task was to find the variables which (in retros-
pect) best discriminated the dangerous from the safe patients at
follow-up. Because outcome variables were highly skewed, as in low
base rate behaviors, initial analyses involved a number of statistical
70


approaches. Discriminant function analysis (Klassen & OConnor,
1988; Klecka, 1984; Steadman & Morrissey, 1981) was selected as the
most appropriate statistic for the data set. A similar analysis for
distinguishing the groups opposed for release by a prediction of
danger was applied.
The stepwise method for deriving the equation, which interac-
tively discriminates the classifications of dangerous patients, elimi-
nates extraneous variables. An F-ratio of 1.00 was selected to be
broadly inclusive and to identify variables with minimal explanation
for differences between SAFE and DANGEROUS groups.
A secondary goal was to determine if the patients predicted by
clinical and agency officials to be dangerous, that is, the group
opposed for release, could also be distinguished by other items of
criminal or psychiatric disruption in history. Items were entered in
logical combination as 1) sociodemographic variables of education,
age, ethnicity, employment, and marital history; 2) clinical variables
of diagnosis, program, alcohol and drug abuse, security, passes, and
the like; and 3) criminal variables of juvenile record, prior violent
crime, incarceration, use of a weapon, and probation history.
Discriminant analyses were first applied separately to the three
categories of items to determine the optimal discrimination of
outcome groups and prediction groups. The wide range of time with
opportunity to recidivate or relapse was controlled for by entering
each subjects release date subtracted from the date of followup.
The six year range for releases was expected to contribute significant
differences in rates of disruption.
71


Variable reduction in the final discriminant function equation
selected the most efficient number and type of variables for distin-
guishing the dangerous from the safe. Confidence levels vary,
according to the specific technique, but in general have an upper
limit of .05 for consideration in predictive equations. Relative
Improvement Over Chance (RIOC) is a somewhat absolute value in
relationship to the distribution being analyzed for predictive accuracy,
so the critical region is positive and directional. RIOC findings with
this sample ranged from a negative (the information was worse than
chance) to an 80 % improvement over chance.
The end product of the various statistical approaches were two
sets of tables, one pair for the dependent variable of DANGER and
one pair for the dependent variable of OPPOSE. The pairs of tables
are for comparison of the Bayesian (actual) probability and equal
probability models. The classification results obtained from the
discriminant function for each of the four tables yielded more
precise information about predictor variables than RIOC. Differences
between and among the technical models were evaluated for utility
and value in comparison and replication.
72


CHAPTER 4
RESULTS
Routine nonparametric tests for association among pre-hospital,
in-hospital, and post-hospital disruptions provided the direction for
subsequent application of Relative Improvement Over Chance (RlOC)
and discriminant function analyses with equal and Bayesian probabili-
ties. An operational definition for dangerous behavior after release
was deduced from the following initial analyses of outcome data.
The proposition that agency opposition to release, that is, a classi-
fication as dangerous, would distinguish the subsequently dangerous
was confirmed by three different statistical approaches.
Definition of Dangerous Disruption
Over half (51.4 %) of the sample had no disruptions leading to
arrest or rehospitalization. The social control mechanisms of arrest
and rehospitalization were closely related; that is, if the patient
avoided arrest, he was also likely to avoid rehospitalization. Of 69
not arrested, 56 were also not rehospitalized (X2 = 23.48, df 1, p.
< 0001). The events leading to social control, and the legal disposi-
tions after disruptive events, were analyzed for operational definitions
of dangerous and nondangerous disruptions. In Table 4.1 initial
frequencies of unclassified events provided the foundation for
organizing the data according to study propositions.


Table 4.1. Disruptions Involving Arrest or Rehospitalization
after Conditional Release (n = 109).
Arrested 40 (36.7%)
Rehospitalized 39 (35.8%)
Neither 56 (51.4 %)
Both 26 (23.9 %)
Disruptions
Assaultive
Threats to harm
Off medication
Psychotic relapse
Drug abuse
Crime Category
Crime of violence
Potential violence
Nonviolent crime
No new crime
23 (21.1 %)
21 (19.3 %)
15 (13.8 %)
20 (18.3 %)
23 (21.1 %)
than one disruption.
15 (13.8 %)
12 (11.0 %)
13 (11.9 %)
69 (63.3 %)
109 100 %
Note: S may appear in more
Note: S appears once in most serious crime category.
No homicides or attempts were included in the criminal charges
after release, but there were five sexual assaults, seven aggravated
assaults, and three armed robberies. Potential violence included four
menacing or simple assaults, one arson, four burglaries, and three
property destruction (mischief) charges. Remaining criminal charges
were for trespass, fraudulent use of credit cards, disturbance and the
like. Two were arrested for violations related to eluding police and
reckless driving.
Criminal conduct also included arrests for violation of court
ordered conditions of release, and for minor disturbances associated
with psychotic decompensation. Except when revocation occurred by


declaring the patient again legally dangerous (n=12), only patients
alleged to have committed crimes of violence (sexual assault, agg-
ravated assault, or armed robbery) were classified as dangerous.
The one new NGRI (predicted safe) was also revoked and in the
classification of dangerous. All of the assaults precipitating arrest or
hospital confinement were captured with the definition of dangerous
disruption.
Various crosstabulations and nonparametric tests for association
were applied to the events of disruption and to the dispositions taken
by agents of social control. As with highly correlated arrest and
rehospitalization, mental health dispositions were found to occur in
response to psychotic behaviors. More than one type of disposition
may also have been used in response to the same event. The
univariate frequencies in Table A.2 include such duplications.
Table A.2. Disposition of Disruptive Events Involving Arrest or
Rehospitalization after Conditional Release (n = 109).
Revocation of C.R. 12 (11.0 %)
New NGRI 1 ( 0.9 X)
Incompetent for trial 3 ( 2.0 %)
Civil or Voluntary 27 (24.0 %)
Correctional 27 (24.0 %)
Note: Disruptive Ss (53) account for more than one disposition.
Nearly half of the sample had significant disruption, but not all
disruptive events involved assaultive behavior, nor did serious crimes
of violence always lead to revocation of conditional release. Six
types of disruptive behavior and five types of disposition were
75


analyzed with bivariate distributions. Of 15 crimes of violence, 10
(67 %) were accounted for by Track II patients. Of 12 revocations of
conditional release, 8 (67 %) were accounted for by Track I patients.
Nonpsychotic Track II patients thus were more likely to have
dangerous behavior handled by correctional disposition.
Combining these various events and dispositions led to the
following two-by-three contingency table used for initial descriptive
statistics. Each patient only appears once in the classification of
outcomes, according to his agency recommendation (or opposition) for
release. Column 3, Dangerous Disruption, includes all patients with
crimes of violence and/or revocation of conditional release.
Table A.3. Types of Disruptions Involving Arrest or
Rehospitalization after Conditional Release
by Hospital Prediction (n=109).
No Disruption Nondangerous Disruption Dangerous Disruption
Predicted Safe 43 (55.1 %) 24 (30.8 %) 11 (14.1 %) 78
Predicted Danger 13 (41.9 %) 7 (22.6 %) 11 (35.5 %) 31
56 (51.4 %) 31 (28.4 %) 22 (20.2 %) 109
X2 = 6.30, df 2, p = .0429
In Table 4.3 and for all further statistical analyses, patients
were counted only once in the outcome classification of NO DISRUP-
TION, NONDANGEROUS DISRUPTION, or DANGEROUS DISRUPTION.
76


In attempts to use the most information possible, but in ways that
best enhanced predictive accuracy in low base rate behavior, the
three outcome groups were then dichotomized by combining the NO
DANGER and NONDANGEROUS groups as SAFE (n = 87), for RIOC
and discriminant function analysis. This yielded about an 80/20 split
in the dependent variable, driving the remaining calculations.
The 22 patients with DANGEROUS DISRUPTION included 8 (18 %
of 44) revocations from Track I and 4 (6.1 % of 65) from Track II.
Track I had 61 % with no new crimes and Track II had 65 %, not a
significant difference. Crimes of violence were twice as likely for
Track II patients. The chronicity of mental illness was reflected in
higher rates of hospitalization and revocation for Track I, and more
use of correctional dispositions for Track II.
Statistical Analyses
Improvement over chance is a function of the maximum possible
correct classifications and the expected frequencies (random or
chance distribution). In any particular study the value or percent
RIOC depends on the size of the sample, the accuracy rate or
achieved correct, and the relative improvement in accuracy that the
item or score adds. The marginal totals that constrained the initial
analysis were 87 SAFE AND 22 DANGEROUS, or an 80/20 proportion-
al split.
The RIOC statistic is a contingency table analysis of categorical
distributions, so technically the value is not dependent on continuous
measures. Individual variables were each tested by RIOC in two-by-
77


two tables. RIOC was also used to determine the optimal cutting
score for yielding the fewest erroneous classifications (predictions)
with continuous variables. Reduction of the number of variables to
the most parsimonious explanations, and the elimination of redundant
measures, was an objective of all statistical manipulations.
Dangerous Outcomes by RIOC
Results with the dependent variable DANGEROUS DISRUPTION
are shown in Table 4.4 where most variables were included in the
categories of 1) sociodemographic 2) criminal history and 3)
clinical course items. Classifications were similar to the Klassen and
OConnor (1988) discriminant analysis with 67 potential predictors.
The organization of information in Table 4.4 used a format from
Loeber and Dishion (1983) to emphasize the errors (false negatives
and false positives) that the model tries to minimize. Percent of the
base rate (T. BR) is the actual rate of dangerous behavior; percent of
the selection ratio (% SR) represents the proportion of the total
sample who would have to be detained to prevent dangerous behavior,
that is, classified as high risk on the independent (predictor) variable.
Individual items at a continuous level of measurement were
analyzed by RIOC to determine the best cutting scores for high and
low risk group assignments reported in Tables 4.4 and 4.5. More than
2 COURT leaves, less than 2 PASSES, any BMT (behavioral manage-
ment transfer), and any hospital ESCAPE were found by RIOC to
improve predictive power significantly. These in-hospital disruptions
were used in unweighted combination as a composite variable (REBEL)
in subsequent discriminant analysis.
78


Table 4.4 Accuracy of Predictions of Dangerous Behavior After Conditional Release (n=109).
False False
True Pos. Positives n % True Neg. Negatives n % % Corr. %BR %SR % RIOC
Sociodemographic
Age < 30 yrs. 10 26 23.8 61 12 11.0 65.1 20.2 33.0 36.4
Minority 12 26 23.8 61 10 9.1 67.0 20.2 34.9 28.6
Educ. < h.s. 13 37 33.9 50 9 8.3 57.8 20.2 45.9 25.0
Never Married 10 39 35.8 48 12 11.0 53.2 20.2 45.0 0.0
Longest Job < 3 yrs. 20 61 55.9 26 2 1.8 42.2 20.2 74.3 66.7
Criminal History
Juvenile Record 14 34 31.1 53 8 7.3 61.5 20.2 44.0 33.3
Prior Violence 10 22 20.2 65 12 11.0 68.8 20.2 29.4 25.0
Incarceration 13 28 25.7 59 9 8.3 66.1 20.2 37.6 35.7
Violent Instant Off. 13 53 48.6 34 9 8.3 56.9 20.2 60.6 35.7
Use Alcohol/Drugs 14 33 30.3 54 8 7.3 62.4 20.2 43.1 38.5
LDS > 5 15 40 36.7 47 7 6.4 56.9 20.2 50.5 36.4
Weapon Instant Off. 10 55 50.4 32 12 11.0 38.5 20.2 59.6 -33.3
Clinical Course
Nonpsychotic Dx 15 50 45.9 37 7 6.4 47.7 20.2 59.6 22.2
High Security 11 18 16.5 69 11 10.1 73.4 20.2 26.6 31.3
Escapes 11 33 30.3 54 11 10.1 59.6 20.2 40.4 15.4
> 25% Lockup 21 63 57.8 24 1 0.9 41.3 20.2 77.1 80.0
> 2 Court Trips 15 41 37.6 46 7 6.4 56.0 20.2 51.4 36.4
< 2 Passes 17 45 41.3 42 5 4.5 54.1 20.2 56.9 44.4
BM Transfers 16 46 42.2 41 6 5.5 52.3 20.2 56.9 33.3
Rebel Score > 2 18 48 44.0 39 4 3.7 52.3 20.2 60.6 55.6
Note: BR = % Baserate; SR = % Selection Ratio; RIOC = Relative Improvement Over Chance.


Table 4.5. Hospital Recommendations For or Against Release by In-Hospital
Disruptions as Predictors (n=109).
False False
True Positives True Negatives % %
Disruptions Pos. n % Neg. n % Corr. %BR %SR RIOC
More than 2 Court Trips 22 34 31.2 44 9 8.3 60.5 28.4 51.4 40.0
More than 25% Lockup 28 56 51.4 22 3 2.8 45.9 28.4 77.0 57.1
BM Transfers 23 39 35.8 39 8 7.3 56.9 28.4 56.9 38.5
Less than 2 Passes 24 31 28.4 47 7 6.4 65.1 28.4 50.5 53.3
Program II 24 41 37.6 37 7 6.4 55.9 28.4 59.6 46.2
Nonpsychotic 18 26 23.8 52 13 11.9 64.2 28.4 40.3 27.8
Escapes 19 25 22.9 53 12 11.0 66.1 28.4 40.4 33.3
High Security 19 10 9.2 68 12 11.0 79.8 28.4 26.6 52.4
SR = % Selection Ratio; RIOC = Relative Improvement Over Chance.
Note: BR = Baserate;


Sociodemographic variables, when measured by RIOC, were
directional, but not at particularly high accuracy rates. Minority
groups had more trouble after release, but were not more likely to be
opposed for release. A stable job history was a low risk indicator,
with two false negatives, but the selection rate classified nearly
three-fourths (% SR) as high risk for DANGER. Marital status was
not a helpful discriminator when used alone against dangerous
outcomes.
Knowledge of a patients prior violent crime improved the
accuracy with an RIOC of 25% and a total accuracy rate of 68.8% .
The number of false positives was reduced to 22, but false negatives
were 12 (11 % is the highest rate of false negatives reported). Prior
violence and other criminal items were subsequently tested in a
composite variable by discriminant function analysis. Clinical
variables helped differentiate the dangerous by accuracy rates and
RIOC, particularly items related to in-hospital security classifica-
tions. Patients released from maximum or medium security had a
greater probability of DANGEROUS outcomes than patients released
from intermediate or minimum security.
The proportion of the length of stay spent on maximum and
medium security wards was calculated in a variable named LOCKUP.
It was first treated as a continuous variable for RIOC analysis of a
cutting score. LOCKUP greater than 25% captured all but one
dangerous outcome and yielded an 80% RIOC, but at the expense of
63 false positives. Patient disruption during hospitalization was
operationalized by the REBEL composite variable, which yielded only
81


four false negatives and produced an RIOC of 55.6 percent better
than chance on the DANGER outcome variable. Disruption in hospital
was also found in the next analysis to be significantly related to
hospital opposition.
Agency Opposition by RIOC
Patients opposed (31) and patients recommended for release (78)
constituted the dependent variable for testing the second research
proposition with results in Table A. 5. No significant differences were
found for OPPOSITION on the variables of ethnic minority, education,
age, marital, job or criminal history.
In-hospital disruptions started to appear to be important to the
clinical prediction variable when using RIOC to analyze patients
conformity to the graduated treatment program. The idea of "going
through the system" came to be operationalized by the absense of
regressive transfers, hospital escapes, high security classification at
the time of release, and failure to have been tested with passes in
the community before conditional release. Individual items were
related, and summing them in a variable called SYSTEM, led to
enough improvement over chance to include the items in subsequent
analyses.
To code for directional purposes in tabulating, the SYSTEM
variable was transformed to one called REBEL, believed to describe
nonconformity to agency expectations. REBEL distributions allowed
False negatives to be minimized to one with low REBEL scores, at
the expense of 62 false positives. Raising the score to over 2
optimized errors to 5 false negatives and only AO false positives. The
82


fewest number of erroneous predictions of safety (recommendation for
release) in this scheme (3) came from percent hospital career in
LOCKUP, but again resulted in a 77% selection ratio and 56 false
positives. LOCKUP was the proportion of STAY spent by the patient
on high security level wards. All values of RIOC are relative and
dependent on the 28% base rate of OPPOSITION in the actual agency
classification of this sample.
Discriminant Function Analysis of Dangerous Outcomes
To evaluate the proposition that civil libertarian models would
work to minimize false positives, a Bayesian model of actual probabi-
lities was derived from the sets of sociodemographic variables,
criminal history items, and clinical variables. The stepwise dis-
criminant procedure eliminated a number of variables from the
equation. Removed were AGE, STAY, RISK, LDS, DIAGNOSIS,
WEAPON, INSTANT OFFENSE, ALCOHOL, DRUG, EDUCATION,
MARITAL, and REBEL.
The first pair of comparisons of the two probability models,
included the same six variables in the final two equations and
discriminant functions. In Table 4.6 the discriminant function scores
assigned group membership(s) in the "Predicted" columns. Significant
variables and their standardized canonical coefficients for distinguish-
ing DANGER were LOCKUP .87; MINORITY .34; PROGRAM -.47; JOB
.30; PRIOR .24; and OPPOSED .22.
In the first probability model (50:50), 35 are classified as high
risk for danger, and about half of those are correctly captured with
the predictor items in the discriminant function. As well, the
83


scheme captures 81.8% of the 22 actual dangerous outcomes. In the
Bayesian model, only 12 get classified as dangerous, but 67% of those
were dangerous.
Table 4.6. Predicted and Actual Group Membership for Dangerous
Outcomes with Equal and Bayesian Probability Models.
Equal Probability
Bayesian Probability
Predicted
Safe Danger Total
Outcome
Safe 70 17 87
Predicted
Safe Danger Total
Outcome
Safe 83 4 87
Danger _4_______18_______22
Total 74 35 109
% Correct = 80.7% (88/109)
Total Errors = 21
False Pos. = 80.9% (17/21)
False Neg. = 19.1% (4/24)
Danger 14_______8_______22
Total 97 12 109
% Correct = 83.5% (91/109)
Total Errors = 18
False Pos. = 22.2% (4/18)
False Neg. = 77.8% (14/18)
BIOC = 73.3%
RIOC = 60.0%
The proportion of the patients hospital stay spent on high
levels of maximum and medium security (LOCKUP) had the strongest
explanatory power for the variance in outcomes. In combination with
the other five variables that met the threshold criterion for inclusion,
hospital OPPOSITION shared but did not dominate the discriminant
function.
Analyses of errors and the type of errors from the libertarian
and the protectionist perspectives were contrasted in Table 4.6.
Minimization of false positives (4) was best achieved by a Bayesian
model. In contrast, minimization of false negatives (4) was best
84


achieved by an equal probability model favored by protectionists.
True to previous accounts, there is a systematic tradeoff in the
direction of errors when minimizing one type over the other. Using
actual probabilities of 80:20 clearly reduced the libertarian concerns
for unnecessary confinement on the basis of a danger classification.
Discriminant Function Analyses of Opposition
Predictor variables and their canonical coefficients for
inclusion in the equation for OPPOSE were REBEL .99; PROGRAM .38;
ABUSER -38; and JOB .-26, for the equal probability equation. The
discriminators yielded a 78.9 X total accuracy rate. The Bayesian
prior probability model removed JOB from the equation and achieved
an 80.7 % total accuracy rate. As in the DANGER analyses, it is
shown in Table A.7 that when the false positives were minimized to
7, the Bayesian model produced 14 false negatives. Likewise, the
equal probability model would erroneously classify 16 patients as
dangerous in order to reduce false negatives to 7.
Any psychiatric diagnosis of alcohol or drug abuse or depen-
dence was selected as an ABUSER variable and found to distinguish
between patients OPPOSED for release and patients recommended for
release. ABUSER, however, was removed from the equation for
predicting DANGER.
85


Table 4.7. Predicted and Actual Group Membership for Hospital
Opposition with Equal and Bayesian Probability Models.
Equal Probability
Predicted No Yes Total
Oppose No 62 16 78
Yes 7 24 31
Total 69 40 109
% Correct = 78.9% (86/109)
Total Errors = 23
False Pos. = 69.6% (16/23)
False Neg. = 30.4% (7/23)
RIOC = 65.0%
Bayesian Probability
Predicted
No Yes Total
Oppose
No 71 7 78
Yes 14 17 31
Total 85 24 109
% Correct = 80.7% (88/109)
Total Errors = 21
False Pos. = 33.3% (7/21)
False Neg. = 66.7% (14/21)
RIOC = 58.8%
Summary of Findings
The PROGRAM variable was included in the final discriminant
function analysis for both DANGER and HOSPITAL OPPOSITION
groups, each group defined as "dangerous" in this study. The
PROGRAM distinction was found to be somewhat of a composite
variable in itself. Significant differences were found between Tracks
I and II in diagnosis, marital, educational, criminal record, alcohol
and drug abuse, and in-hospital disruptions, described under the
Sample. The shared variance of those variables may weir have been
captured by PROGRAM, explaining why a number of variables thought
to be important were removed by earlier discriminant analyses.
The LOCKUP variable was also correlated with PROGRAM.
Track II patients spent proportionately more time on high security
wards. The underlying construct being measured may have been
86


personality, assessed danger in hospital, "criminality" rather than
abnormal mental condition, nonresponsiveness to treatment, or none
of the above. Whatever clinical or management criteria contributed
to program assignment probably contributed to the operational
definitions of dangerous behavior after release, as well as the
classification of danger by agency opposition to release. Subsequent
decisions for revocation (12) or insanity (1) were often made by the
same agencys clinical administrators and psychiatrists.
The REBEL composite measure of in-hospital disruption (court
trips, behavioral management transfers, escapes, and no passes) was
strongly associated with agency OPPOSITION. Surprisingly, however,
in-hospital disruption was not predictive of subsequent DANGEROUS
DISRUPTION. The REBEL variable was removed from the equations
for DANGER under both prior probability conditions and did not
discriminate the dangerous from the safe groups. And finally,
OPPOSITION did enter the prediction of DANGER classification, but
not weighted as powerfully as the other five discriminators.
The effects of experimentally omitting one variable and then
another from the discriminant function confirmed the value of the
technique of combining variables to explain the most variance. When
the OPPOSE prediction included PROGRAM, REBEL, AND ABUSER,
the total correct classification rate was 70.9 % under equal probabi-
lity conditions. Removing PROGRAM reduced accuracy to 71.6 %.
Including PROGRAM and removing REBEL dropped accuracy to 55.9 %
and the discriminant function also removed ABUSER. It would appear
that if decision makers had only one item of information to make
87


recommendations to the court, disruption in hospital (REBEL) would
be the most helpful. If evaluators of those predictions had only one
item of information to identify the DANGEROUS after release,
security classification in hospital (LOCKUP) would be most helpful.
68


CHAPTER 5
DISCUSSION
The simplistic notion of first generation research was to predict
everyone safe in light of the high rate of erroneous predictions of
danger. The idea has some statistical substance as well. If all 109
patients had been predicted safe, the error rate would have been the
22 (20%) who were subsequently dangerous. In actual fact, all 109
NGRIs in this sample were indeed predicted safe by someone,
including the judges and juries who released 31 patients over the
objections of the agency.
As in the Colorado (1977) study of 111 NGRIs also predicted to
be safe, the present sample had significant trouble after release in
nearly half of the cases. Arrest and rehospitalization were defined
as external controls exercised in response to social disruption. The
base rate for this sample is 48.6 percent when both nondangerous and
dangerous disruptions are included. The further distinction made here
to define only 22 as dangerous relied on the legal definition for
revocation, and on FBI Index crimes of violence.
Public protectionists have scientific evidence from this study,
and Steadman and Morrissey (1981), that an equal probability model is
one way to best minimize false negatives. No combination of
variables attempted here managed to reduce false negatives to zero,
outside of predicting every patient to be dangerous (a false positive
ratio of 87/22 and total accuracy of only 20%). The best cases for a


Bayesian model consistently reduce false positives at the expense of
false negatives of 10, 12, or more.
Caution is suggested before cross validation studies can attempt
replication of specific predictor variables. As pointed out by Klassen
and OConnor (1988) these statistical techniques produce "tailor-made"
equations for each particular set of data. There is thus an inherent
instability in low base rate behavior with relatively small samples.
On the other hand, it may be helpful to have program-specific and
sample-specific indicators for risk assignment of future NGRI releases
from this and similar agencies.
Uncertainty about what PROGRAM assignment is actually
measuring is acknowledged, but whatever it may be, treatment
assignment is predictive here of both hospital OPPOSITION to release
and DANGEROUS DISRUPTION after release. In-hospital disruption
as measured by behavioral management transfers, escapes, high
security at release, and no passes (REBEL) assured that the agency
would oppose release on the danger criterion.
Perhaps level of social functioning (premorbid life style) and
level of functioning in the hospital are related to treatability, rather
than to dangerous characteristics. Some may simply start out in a
low risk group with less need for treatment than the highly danger-
ous. They may proceed without disruption through the graduated
security program, using passes for community re-entry, and meeting
SYSTEM expectations. They may not be REBELS, but they may not
have been dangerous either. As well, institutions are notorious for
exaggerating the very behavior they are designed to control (Sykes,
90


Goffman, Toch). If some patients start out as REBELS, or their
experience in program contributes to becoming REBELS, these
findings suggest they are unlikely to be released with the hospital's
blessing.
The REBEL variable did not remain in the discriminant function
analyses for dangerous outcomes, being replaced with a different
combination of predictor variables (ABUSER, JOB). One explanation
is that OPPOSE itself may be a function of the items that were used
in composite variables. Thus when OPPOSE entered the prediction
formula, individual distributions on the variables became rearranged
for the optimal discrimination of the safe and the dangerous. The
strength of LOCKUP for DANGEROUS DISRUPTION may also have
some colinear relationship with items in REBEL.
Base Rates and Probability Models
This base rate approaches the equal probability model that
Gordon (1982) suggests is reasonable for high risk, homogeneous
groups of known offenders. He contends that the Koppin and Kozel
samples, and his Patuxent group, were much higher risk than
ordinary citizens. His (1977) solution to high rates of false positives
was to add 111 nuns (expected safe outcomes) to the Colorado
sample, a humorous lesson in heterogenity. Gordon says that
whenever material becomes more homogeneous, corrrelations are
lowered. He asks "Why should dangerousness be an exception?"
(1982, p. 298).
Gordon continues the analysis by pointing out the seeming
paradox that as predictive values approach the mid-range probability
91


of .50, outcomes actually become more heterogeneous. The propor-
tional probabilities set for discriminant analysis, and the marginal
(actual) distributions used in RIOC, were attempts to account for
skewed outcomes. And in Gordons review of the few empirical
studies, and in this present research, outcome criteria for danger are
dichotomous or zero-one probabilities. The classification is acceptable
after the fact; but when predicted, danger is not an all-or-none
probability of 0 or 100 percent.
The selection ratios calculated with various high risk indicators
constitute a statement of sample probability, not the same as saying
that an individual member of the high risk group has a precise
probability of being dangerous. In this sample of 109, hardly anyone
had a zero probability, as with the offender populations Gordon
studied in his policy analysis. On the other hand, when making
decisions to release or to detain, the decision does become dichoto-
mous. The offender either is conditionally released (or paroled, or
allowed bail, etc.) or he is confined as dangerous. Likewise, until
scales or schemes for distinguishing severity of dangerous outcomes
are developed, he is either dangerous or safe at follow-up.
Explicit and Replicable Criteria
One of the Morse (1983) objectives for further scientific study
called for more precise and validated criteria for classifying danger.
Discriminant function analyses probably is the best technique used so
far in the prediction literature, to deal with both kinds of error,
and the danger-safe dichotomy. A broad definition of danger was
used in the revocation of conditional release in a few cases here.
92


Disruptive events resulting in other methods of rehospitalization,
classified here as nondangerous, did include criminal conduct that
could be defined as dangerous in a broader definition.
Psychiatric relapse in a chronic schizophrenic, known to be
assaultive when off medication, was declared to be imminently
dangerous in 27 cases of civil rehospitalization. Some of those cases
were classified here as NONDANGEROUS DISRUPTION. The defini-
tion for the 22 DANGEROUS is straightforward and replicable by
other states with similar statutes (California, Illinois, Oregon). The
other kinds of disruption are probably less well-defined and include
both criminal recidivism and psychiatric relapse. Using arrest and or
rehospitalization for 24 hours or more does help operationalize.
Increased Degree and Probability of Harm
The second Morse objective called for research to substantiate
preventive confinement on the basis of danger predictions. The
concern reflects the libertarian concern for false positives. Several
cases from that cell were researched and found to have had nondan-
gerous disruption, suggesting that they were not totally misclassified.
The best ratios found here of false positives to true positives were in
the range of 1.6 to 1.0. Bayesian models yielded one contingency
with 85 % true positives (accurate identification of danger) but at the
expense of 16 false negatives.
The Floud and Young (1982) analysis took a different perspec-
tive, and discussed the validity of retributive justice and the fact
that samples are known to have been responsible for dangerous
behavior before being considered for high risk classification. Their
93


Full Text

PAGE 1

PREDICTION OF DANGEROUS BEHAVIOR FOR RELEASE OF THE CRIMINALLY INSANE IN COLORADO by Mary Katherine Koppin B.S., Southern Colorado State College, 1970 M.A., University of Colorado, 1974 A thesis submitted to the Faculty of the Graduate School of the University of Colorado in partial fulfillment of the requirements for the degree of Doctor of Philosophy Graduate School of Public Affairs 1990

PAGE 2

This thesis for the Doctor of Philosophy degree by Mary Katherine Koppin has been approved for the Graduate School of Public Affairs by Date

PAGE 3

Koppin, Mary Katherine (Ph.D., Public Administration) Prediction of Dangerous Behavior for Release of the Criminally Insane in Colorado Thesis directed by Professor Eric D. Poole ABSTRACT The public policy of preventive detention of the mentally ill is legitimized by insanity statutes in Colorado. Not Guilty by Reason of Insanity (NGRI) cases were followed after conditional release from the state's only psychiatric security hospital. Agency release recommendations, made on a danger products test, were evaluated for predictive accuracy by Relative Improvement Over Chance (RIOC). Disruptions before and during hospital career were tested for predictive relationships with criminal recidivism, rehospitalization, and revocation of conditional release. Discriminant analysis was used to determine which variables were useful in predicting membership in risk and outcome groups. Clinical predictions were found to distinguish NGRis who were subsequently dangerous. Crime of commitment, youth, and length of stay were not predictive of dangerous outcomes. Progression through a treatment program of gradual reduction in external controls was correlated with clinical predictions and with outcome. Conformity to the program resulted in favorable recommendations for release and fewer dangerous behaviors after release. Diagnostic and program differences within the sample of 109 NGRis were suggested to reflect two distinct populations. Signed

PAGE 4

CHAPTER !.INTRODUCTION CONTENTS The Research Problem Comparability of Samples Replication of Methods Generalizing from the Findings The Public Policy Problem . Legal Background Clinical Background Methodological Background Scope of the Problem Purpose of the Present Study Organization of the Dissertation 2.REVIEW OF THE RELEVANT LITERATURE Legal Foundations . Jones v. United States Legal Definitions of Danger Outcome Criteria for Danger Psychological Assessment of Danger Personality Scales The Bridgewater Legacy . . Psychopathy and Validity of Diagnosis Prediction Accuracy 1 2 3 4 5 6 8 13 14 16 19 19 22 23 24 26 28 30 31 33 34 36

PAGE 5

The Baxstrom and Dixon Cases The Legal Dangerousness Scale Political Predictions and Dixon Cases Prediction Accuracy Research Propositions 3.METHODS . . . . . . . . . . Design and Prediction Models Selection of the Sample The Treatment Setting . Program Description Prerelease Testing in the Community Maximum Security Programs Description of the Sample . Demographic Characteristics Length of Hospital Stay and Crime Release Procedures The Legal Release Criteria Patient Petition for Release Psychiatric Recommendations The Administrative Review . The Official Agency Recommendation Adversarial Release Hearings Data Sources and Limitations Patient Data System Hospital Records Level of Measurement v 37 39 41 42 43 45 46 48 49 50 52 53 53 54 55 56 56 57 58 58 59 60 61 62 62 63

PAGE 6

Validity and Reliability o Operational Definitions o o o Disruption After Release o Disposition After Disruption o Prehospital Disruptions In-Hospital Disruptions Statistics 4oRESULTS Definition of Dangerous Disruption Statistical Analyses Dangerous Outcomes by RIOC o Agency Opposition by RIOC Discriminant Function Analysis of Dangerous Outcomes Discriminant Function Analyses of Opposition o Summary of Findings o 5oDISCUSSION o Base Rates and Probability Models Explicit and Replicable Criteria o Increased Degree and Probability of Harm o Research Relevant Evidence o o o Programs to Reduce Propensity for Violence o Provision of Thorough Review of Confinement BIBLIOGRAPHY o o o o o vi 63 64 65 66 66 68 69 73 73 77 78 82 83 85 86 89 91 92 93 94 95 96 99

PAGE 7

TABLES 3.1. Crime of NGRI Commitment by Length of Hospital Stay in Months: Conditional Releases 1980-1986. (n=109) 55 4.1. Disruptions Involving Arrest or Rehospitalization after Conditional Release (n = 109). . . . 74 4.2. Disposition of Disruptive Events Involving Arrest or Rehospitalization after Conditional Release (n = 109). 75 4.3. Types of Disruptions Involving Arrest or Rehospitalization after Conditional Release by Hospital Prediction (n=109). ... 76 4.4. Accuracy of Predictions of Dangerous Behavior After Conditional Release (n=109). . . 79 4.5. Hospital Recommendations For or Against Release by In-Hospital Disruptions as Predictors (n=109). 80 4.6. Predicted and Actual Group Membership for Dangerous Outcomes with Equal and Bayesian Probability Models. 84 4.7. Predicted and Actual Group Membership for Hospital Opposition with Equal and Bayesian Probability Models. 86 vii

PAGE 8

CHAPTER 1 INTRODUCTION The public policy of preventive detention of the criminally insane is based on an assumption that psychiatrists and other mental health professionals can distinguish dangerous from safe persons. There are few empirical investigations of danger prediction, and most researchers concluded that psychiatric opinions about future danger-ousness were usually wrong (Ennis & Litwack, 1974; Monahan, 1984; Steadman & Cocozza, 1974; Thornberry & Jacoby, 1979). The lay public believed psychiatrists could predict violence; the research findings suggested they could not; but the law required that they must predict. Psychologist and attorney Stephen Morse (1983) recognized the legal mandate and recommended the use of expert evidence with improved social science: 1. to make the criteria more explicit, 2. to increase the degree and probability of harm required, 3. to admit only that evidence that research demonstrates is relevant. 4. to offer programs to those confined that might reduce their propensity for violence, 5. and to provide thorough periodic review of the continuing need for confinement. (p. 18) Morse's objectives were formulated in light of the consistent finding that clinical predictions were accurate in only one third of the cases.

PAGE 9

Statistical predictions achieved better accuracy, but relied on arbitrary social predictors (age, sex, race). The literature on the prediction of danger is primarily the work of a small number of public policy critics who reported inaccurate risk assignment by psychiatrists. The direction of error was consistently reported as one of overclassifying mentally ill offenders as dangerous, whether by clinical or statistical assignment of risk (Pfhol, 1978; Steadman & Morrissey, 1981). The infrequency of violent events was one reason for low accuracy rates. Rare events and ra..11.dom probability were considered in the predictive analyses of Loeber and Dishion (1983) in Oregon delinquency studies, a statistic applied here to an adult sample of criminally insane patients. Most investigators measured prediction accuracy in retrospective studies by classifying certain outcome behaviors as dangerous, and then searching for offender characteristics that might have been predictive of outcome. Arrest for crimes of violence and/or rehospitalization for assaultive behavior were defined as dangerous recidivism in nearly all of these core studies. The Research Problem Because most offenders, including those who are criminally insane, remain confined when classified as dangerous, researchers seldom have opportunity to test the accuracy of predictions. Dangerous mental patients are detained in prisons or hospitals without the freedom to test their danger classification. Courts interrupted such continued confinement when they ordered the 2

PAGE 10

release of several hundred patients in maximum security hospitals of New York in Baxstrom v. Herold (1966) and Pennsylvania in Dixon v. Commonwealth (1971). Due process rulings provided the first opportunities for follow-up of presumably dangerous and insane criminals. The court's action was acclaimed by social scientists as an opportunity for a natural experiment, a study of the public policy of the preventive detention of mentally ill offenders. Comparability of Samples Analysts broadly defined the BaXstrom and Dixon populations as criminally insane, generalizing from mentally ill prisoners, pretrial incompetents, sexual psychopaths, and patients unmanageable in civil hospitals. Comparability among research samples and distinctions between mental health and correctional populations were initial methodological problems noted by researchers. Baxstrom and Dixon release samples were primarily prison inmates who became ill or troublesome and were transferred to security hospitals. Neither sample was specifically predicted to be dangerous (Greenland, 1985) nor had replicable variables been used to classify the patients as dangerous at follow-up. The technical definition of criminal insanity includes only criminal defendants who have been acquitted of a crime by being found Not Guilty by Reason of Insanity (NGRI). NGRI defendants have been relieved of criminal responsibility by avoiding culpability and accountability before the law, unlike the convicted criminal who may be punished. The acquittal makes any subsequent criminal sanctions for detention of the NGRI conceptually quite different 3

PAGE 11

from sanctions applied to convicted and sentenced correctional populations. Comparison of samples, replication of methods, and generalization of findings are as relevant to the research problem as the theoretical issues of public policy and preventive detention. Replication of Methods Dangerousness is often assumed by clinicians to be a characteristic or a trait within the individual, a propensity that can be diagnosed. No adequate typology of violent persons exists, however, according to Mulvey and Lidz (1985). No psychological tests directly measure such a propensity, although indirect measures of hostility, aggression, and psychopathy have shown promise in Canadian studies (Hart, Kropp, & Hare, 1988). Persons who scored high on the HARE Checklist of Psychopathy (PCL) had higher rates of criminal recidiv-ism and revocation of conditional release than nonpsychopaths and other lower scoring subjects. It is not known, however, how much dangerous behavior can be accounted for by psychopaths. The following work includes a sample of chronically mentally ill (the majority were nonpsychopathic) patients. Base rates of psychopathy within samples of hospitalized offenders may differ among jurisdictions according to commitment laws. Prediction of danger studies, typically retrospective and exploratory, have generally not used instruments that have been normed or validated on other samples. New York scholars Cocozza and Steadman (1974) did produce a four item "scale" which was a summary of criminal history in binary code. The New York Legal Dangerousness Scale yielded strikingly similar results on a sample of Dixon 4

PAGE 12

patients, and with a sample of younger and less institutionalized NGRI patients released in Colorado (Koppin, 1977). Predictive accuracy, base rates of recidivism, and criminal histories were also quite similar in Gordon's (1982) comparisons of Maryland patients with New York, Massachusetts, and Colorado samples. The primary problem with pre-commitment variables of criminal record as predictors is their fixed and permanent nature. Only growing older will take an offender out of the high risk group. Nothing in treatment can change his criminal history as fixed predictors of danger. Generalizing from the Findings The reported findings of inaccuracy and over prediction of danger led to protectionist criticisms in the literature. Conservatives charged libertarians with generalizing from an opportunistic case mix, aged populations, convicted inmates and patients who were never charged with crimes, and imprecise definitions. Spokesman Saleem Shah of The National Institute of Mental Health led public policy analysis with his proposals for a research paradigm (1975, 1978, 1981, 1986). He recommended distinctions for the major concepts of "dangerous" behavior, violence, and "dangerousness." The terms had various uses and definitions in the law and in the psychological literature. The early studies formed what Monahan (1984) called the first generation of prediction research with mentally ill criminals. The works defined the state of the art of psychiatric prediction and the empirical study of relevant populations. Forensic psychiatrists and psychologists were following the Baxstrom and Dixon results, along 5

PAGE 13

with the clinical predictions reported by Kozel, Boucher, and Garofalo (1972) on a sample of sexual psychopaths from a Massachusetts security hospital. The Kozel findings supported a treatment model and were more acceptable to clinicians than the New York and Pennsylvania reports of psychiatric inability to diagnose and predict. Considerable influence on public policy was attributed to the early studies, with technical criticisms receiving less attention in the policy arena (Klassen & O'Connor, 1988). Conclusions that clinical predictions were most often wrong may have been generalized from insufficient evidence and faulty methods (Floud & Young, 1982; Gordon, 1977, 1982; Mulvey & Lidz, 1985). Illusory correlations and problems of colinearity were alleged by these and other authors. Experts in probability theory said that not enough information was known about base rates, or the frequency of dangerous behavior among particular populations, to have developed sophisticated probability theory or models. Because most items reported to be associated with dangerous behavior were at a categorical (nominal) level of measurement, parametric statistics were rarely appropriate. The Public Policy Problem Libertarian scholars concluded that society locked up far too many innocent persons under the guise of danger predictions. Inaccurate classification of persons as dangerous who are actually safe leads to costly and unnecessary utilization of hospital and prison resources, according to the libertarian perspective. In a 6

PAGE 14

medical or clinical model, the false positive is judging a heal thy person to be sick and proceeding with a treatment. This is considered to be preferable to the false negative of misdiagnosing health and failing to treat. The policy issue for civil libertarians is one of false positives and unnecessary treatment and confinement. Monahan (1973, 1981) was representative of social theorists who used primary data generated by Baxstrom and Dixon cases in public policy debate. Policy analysts monitored reports on the 1,000 court ordered releases from New York and Pennsylvania security hospitals. Findings were that criminal patients were not very dangerous and behaved much like institutionalized civil patients in community placements (Steadman & Cocozza, 1974; Steadman & Keveles, 1972; Thorn berry & Jacoby, 1979). Conservative practitioners (Deitz, 1979; Floud & Young, 1982; Kozel et al., 1972) argued in favor of social control and public protectionist policies. Psychiatrists using a danger prediction model are said to make conservative predictions and retain for treatment, rather than release the patient and risk a false negative. When a person is predicted dangerous and is simply confined for treatment, there is less political and media impact than when a person who was predicted to be safe commits a dangerous act. Viewed from the perspective of public protectionists the policy issue is one of false negatives. The public administrators at the study hospital for this research described their mission as protectionist with conservative policies to minimize false negatives (see Appendix). The avoidance of critical media coverage is also implied 7

PAGE 15

from policies of the Governor's office. Erroneous prediction of safety may be followed by hasty legislation passed in response to a violent act by a former mental patient. Subsequent restrictive changes in public policy may proceed in the absence of rational information (Mulvey & Hilz, 1985). An emphasis on civil liberties leads to a focus on the error of over prediction of danger, whereas an emphasis on public protection leads to a focus on the error of inaccurate prediction of safety. The right to treatment in the least restrictive environment, as well as constitutional rights to due process, are premises from the libertarian perspective. The responsibility to protect the public, and to avoid liability for the dangerous behavior of psychiatric patients, are premises from the protectionist perspective. Legal Background Although there was little scientific evidence that mental health professionals could accurately predict individual events of dangerous behavior, federal and Colorado laws continued to require release decisions based on psychiatric assessment of the likelihood of dangerous behavior. Eligibility for release from an NGRI commitment in Colorado (CRS 1983, 16-8-120) requires a psychiatric opinion that the patient "no longer suffers from an abnormal mental condition that is likely to cause him to be dangerous to himself, to others, or to the community in the reasonably foreseeable future." The problem of long-term prediction of danger for the criminally insane contrasts with the short-term (civil) assessment of imminent danger. The legal criterion for condi tiona! release required the 8

PAGE 16

psychiatrist to relate abnormal mental conditions to a propensity for future dangerous behavior. The statutes do not provide a definition for foreseeable future; in clinical practice the treatment plan usually required a minimum of one year on conditional release. In the NGRI sample it was not unusual for psychiatrists to report that the schizophrenic would remain not dangerous if he continued neuroleptic medication, or that the alcoholic might again become dangerous if he were to drink. Psychiatric assessment and testimony in criminal insanity cases thus takes the form of clinical prediction. Clinicians recommended restrictions on gun ownership, use of substances, and compliance with prescribed medications, which were used as court ordered conditions for remaining on release status. Actuarial or statistical predictions are seldom used in decisions about individuals. Scales or instruments are more often a research tool for retrospectively applying measures to groups of offenders after clinical predictions are implemented and follow-up data are obtained. Actual application of scales or measures that have no norms or validity would be inconsistent with civil rights legislation, and with the ethics of The American Psychological AssociationJs standards for tests and measurements (1985). Statistical associations with dangerous behavior have included demographic items of youth, sex, race, juvenile record, or number of arrests and incarcerations. Actuarial models do not rely on causal theories, but items are used to show that increased probability of dangerous behavior occurs with increased risk factors that assign offenders to risk prediction groups. An ideal situation for court 9

PAGE 17

work would be to state an individual's probability as a member of a particular high (or low) risk group, recognizing the probability as only a base rate for some known group that may be similar to the individual. The Colorado release test is a type of "product" test. Nevertheless, clinical practice observes dangerous persons who are not mentally ill, and many mentally ill persons who are never dangerous. Causality (that is, the abnormal mental condition causes the dangerous act) is implied in a number of commitment statutes. In the New Hampshire insanity test criminal behavior is considered the "product" of mental disease or defect (Keilitz & Fulton, 1984). Disturbed or criminal behaviors were thought to be the result of hallucinations, delusions, or disordered emotions that came to be legally defined as mental disease or defect. The United States Courts used a product test for a generation, associated with the case of Durham v. United States, 1954). The. Durham product test was cumbersome, alleged to encourage wide psychiatric speculation, and was abandoned in favor of the American Law Institute's (ALI) Model Penal Code. The ALI standard was a test of "substantial capacity to appreciate the criminality of his conduct or to conform his conduct to the requirements of law" (Keilitz & Fulton, 1984). The legal concept of capacity was an insanity test more tolerable to psychia-trists than the concept that implied causality. Not until legal reforms in 1983 (and only for NGRis whose crimes were commi ted after July 1, 1983) did Colorado adopt language similar to ALI. The legislature retained the danger criterion, similar to a product test, 10

PAGE 18

and added the capacity to appreciate the law in the release test. The burden of proof was on the patient to prove safety and sanity. The Colorado release test for this sample required a psychiatric assessment of an individual's likelihood of dangerousness, as the result of an abnormal mental condition. Dangerous outcomes among 109 conditional releases reported in Chapter 4 were categorized according to behaviors handled by police and correctional authorities, and behaviors that required psychiatric hospitalization. These and other definitional problems with dependent variables were detailed in the review of the relevant clinical and legal literature. Colorado law assumes that psychiatric inpatient hospitalization is necessary for an unspecified period of time for all defendants aqui tted by reason of criminal insanity. The issue of dangerousness, however, is not legally relevant at the time that a Colorado defendant is tried and adjudicated insane. The finding of NGRI for the sample in this study was based on each defendant's "cognitive" ability to know right and wrong, or an inability to refrain from the wrong (the 1843 M'Naghten test and the irresistible impulse). Colorado law interprets the M'Naghten rule by defining insanity as a defect in reason (that is, cognitive ability) due to disease or defect of mind at the time of a criminal act (CRS 1973 16-8-101). The affirmative defense assumes the defendant commited the act and presumes continued insanity. Psychopathy (antisocial or sociopathic disorders) and substance abuse were not excluded during the study period by legal definitions of mental disease or defect. Consequently, the sample included some 11

PAGE 19

patients who resembled correctional cases and others who resembled the chronically mentally ill. The appropriateness of the insanity plea, or the many philosophical forensic arguments about criminal responsibility, are not particularly relevant to the subject of prediction of dangerous behavior of NGRI patients after hospitalization. One case study approach, for the reader interested in the philosophical controversies, is The Trial of John W. Hinckley, Jr. by Low, Jeffries, and Bonnie (1986). Portions of the trial transcripts provide an excellent introduction to diagnostic issues. An NGRI finding is an automatic and indeterminate commitment to a security hospital rather than prison, and the defendant can never be tried again for the crime. The U.S. Supreme Court affirmed the principle of indefinite duration of hospitalization for the criminally insane by their decision in Jones v. United States (1983). Michael Jones, a shoplifter, spent many years in Saint Elizabeth's, a federal security hospital. In response to an equal protection argument, the court ruled that Jones belonged to a special class of NGRI persons, unlike mentally ill persons under ci vii commitment. Jones was contesting his indeterminate length of stay. The danger criterion for release was not specifically addressed by the defendant, prosecutors, or the court. The court did rule, however, on the constitutionality of involuntary and indeterminate confinement in a security hospital for this class of patients. No matter how trivial or serious the instant offense, hospitalization did not constitute punishment and could continue for Jones and several thousand 12

PAGE 20

other NGRis in United States jurisdictions. The spirit of preventive detention was upheld, if not the precise standard of predicted danger. Clinical Background Colorado insanity laws allow a rehabilitation and treatment model. The courts approve gradual increase of patient privileges and reduction of supervision and external controls. This model is supposed to test patient dangerousness, and prepare him for reentry in the community. Patients in the present research were treated in a graduated security level program that provided privilege incentives for progression through certain treatment phases. The taking away of privileges by regressive transfer to a higher security level was hypothesized to be one of several indices of the patient's failure to demonstrate safety by "going through the system." The present work, unlike most of the previous prediction studies, did include variables that might infer treatmep.t effects. The present sample came from two distinctly different programs of rehabilitation, organized around patient characteristics, generally categorized as level of social functioning. Recent Massachusetts work with a typology of sex offenders included the construct of social functioning (Rosenberg et al., 1966). Their classification scheme was similar to the treatment tracking of patients in this sample, diagnosed and assessed according to their premorbid lifesyles. Treatment Track I patients tended to have been psychotic, chronically ill, socially isolated with few verbal skills, prosocial in values, and in need of social learning of daily living skills in a 13

PAGE 21

supportive atmosphere. Track II patients tended to be personality disorders or substance abusers, socially gregarious and verbal, antisocial in values, and in need of character development in a group and community milieu. Program patients were expected to have distinguishable differences in characteristics, treatment disruptions, predictions and dangerous outcomes. Methodological Background Retrospective study of variables that might have had predictive value in most of the core studies found that juvenile record, prior violent crimes, severity of the instant offense, and age were most frequently and significantly associated with subsequent dangerous behavior. Criterion behaviors were assaults and rehospitalization and/or rearrest for serious crimes of violence. A serious problem, however, in practical application of these predictor variables is that risk level is assigned at the time of commitment (record, offense, and age) and is fixed except for growing older. The New York and Pennsylvania samples were each found to be safer if over 50 years of age. If crime is a function of youth and masculinity, then today's generation of offenders would have to be incarcerated for decades. The research here concerned preventive confinement of the criminally insane on the basis of predictions of danger, and concerned the methods and models for analyses of those predictions. Colorado case law reflected a broad definition of dangerousness (a likelihood to become dangerous to oneself, others, or the community) in the foreseeable future. Such a broad definition would encompass all subsequent illegal activity, rather than only seriously assaultive 14

PAGE 22

crimes. A major part of the present analyses focused on the inherent methodological problem of safe/danger prediction dichotomies and safe/danger outcome classifications. The implied zero-one probabilities led to a number of experimental groupings of predictor variables and classifications of behaviors after release. Disruptions after release were classified, as in previous studies, according to the severity of criminal charges, and the action of the courts of commitment in revocations of conditional release. Arrest for any reason and/or rehospitalization under any legal status was defined as disruption. Reasons were then classified according to major and minor disruptions, deducing the subgroup defined as dangerous (n=22) according to the frequencies and distributions of a number of disruptive behaviors. Analyses were described in Chapters 3 and 4, and results were interpreted in Chapter 5. Equal probability and actual probability distributions were derived by discriminant function analyses. Initial analyses began with many potential predictor variables (37) and a goal of variable reduction for the most parsimonious explanations. Items were tested for significant discrimination of dangerous outcomes, and the same variables were tested for differentiating groups defined by hospital predictions of danger. Bayesian probability models consistent with a civil libertarian approach were applied to minimize erroneous predictions of danger. Similar to the contrast made by Steadman and Morrissey (1981), an equal probability model consistent with a public protectionist approach was compared to minimize erroneous predictions of safety. 15

PAGE 23

Discriminant function analyses were used to reduce the number of variables that distinguished dangerous from safe outcome groups, and groups opposed by the hospital for release. Studies with primary data on mentally ill criminal offenders provided the methodological model for the secondary literature and for this research. Except for recent RAND studies which used low, moderate, and high probability group assignments with correctional samples (Conrad, 1985; Greenwood & Turner, 1987), most researchers used a zero-one probability assumption to make retrospective assignments to dangerous or not dangerous risk categories. Those assignments implied either a 100 percent probability or a zero probability to an individual's propensity for danger. Outcome variables of recidivism or relapse were similarly dichotomized. Nearly all studies used this method to report unacceptably high rates of false positives. The zero-one assumptions and previous findings were challenged in the theoretical work of Floud and Young (1982), Gordon (1977, 1982), and Holland, Levi, and Beckett (1981), discussed here in Chapter 5. Scope of the Problem Shah (1978) estimated that 50,000 decisions were made daily in the United States on the basis of offender potential for danger. A daily population of about 80,000 to 100,000 persons were involuntary commitments in mental hospitals (Morse, 1983). Prison security classification, bail and bond eligibility, civil hospitalization, probation, parole, and release from insanity commitment are examples of decisions that are essentially clinical. Common to most such judgments is consideration of the individual offender's criminal record. 16

PAGE 24

When a psychiatric record or violent behavior was also present in the history of California parolees, the likelihood of predicting future dangerous behavior was greater (Wenk et al., 1972). The Wenk studies were the first with large samples and multivariate analyses, but the false positive ratio was quite high. The magnitude of the policy and research problems is affected by the public perception that links violent crime and. mental illness. The spectacular nature of violence, as well as media reporting of a few notorious cases, as in The Denver Post on December 18, 1982, creates strong public opinion about criminal insanity as a defense. The public is disparaging of mental health and criminal justice systems that "allow crazy criminals to walk free." Pasewark, Jeffrey, and Bieber (1987) reported, however, that the NGRI plea was entered in less than one percent of all Colorado criminal cases, and was successful in less than half of those. Although small in absolute numbers, the commitment and release of the criminally insane are perceived to be a public policy problem, one which warrants empirical investigation. In addition to the broader background of research findings, clinical issues, and legal foundations, there was a decidedly political background for public administration and agencies of social control. A political and public policy crisis contributed to the timeliness and appropriateness of this particular study of dangerous behavior among the Colorado insane. John Bromley (1980) was one of many area journalists who indicted police and mental health professionals for not predicting and preventing a rash of homicides in the Denver area. 17

PAGE 25

A number of clients known to mental health agencies were responsible for over a dozen murders within a few months of each other. An Executive Order by Governor Lamrn (April 14, 1980) called for study of the publicized cases and the scientific analyses of relevant target populations for the prevention of violence. A legislative task force recommended reform of conditional release laws for NGRI commi trnents, even though the Denver crimes had not involved any former NGRis or conditional releases. Senate Bill 1, sponsored by Senator Ruth Stockton (R-Lakewood), was passed in the 1981 Session just after the John Hinckley presidential assasination attempt. A Colorado psychiatrist was sued by survivors of the Hinckley shooting for not predicting and preventing his patient's violent behavior. Hinckley's subsequent insanity trial provoked massive legal reform in most states and the federal jurisdictions. These events formed the media background for policy debate. Parallel to the executive's problems in mental health agencies was a class action suit by inmates of the state prisons in Ramos et al. v. Lamm et al. (1979). Classification schemes, crowding, psychological services, and administrative procedures were scrutinized. The federal judge concluded that the old maximum security unit was unfit for human habitation. Judge Kane's 75 page order included ten pages about inadequate health care, with direct criticism of the state hospital's forensic unit. Hospital decisions were influenced by this background of conservative and protectionist policy debate. Purpose of the Present Study 18

PAGE 26

Objective, valid, and replicable methods for assigning individual risk of future danger in forensic populations have yet to be developed. This work retrospectively evaluated agency predictions made for the condi tiona! release of 109 male NGRis in a six year release cohort. Accuracy of clinical predictions was measured by follow up for criminal recidivism or psychiatric rehospitalization as a result of dangerous behavior in the community. Attempts to type the kinds of disruptive behaviors according to severity were similar to the Holland et al. studies and a 15 year longitudinal study of conditionally released patients in Maryland (Spodak et al., 1984), Findings were expected to contribute to a sparse NGRI data base and to provide clinicians with objective feedback about danger assessments. There are few, if any, studies with primary data on danger predictions made for an NGRI sample of younger, less institutionalized patients. Unlike previous studies, specific assessments of likelihood for danger were made by psychiatrists on a legal criterion. Technical methods for evaluating the accuracy of predict-ions were explored. Measures for comparing predictive value to results achieved by chance distribution were taken from delinquency research (Loeber & Dishion, 1983). The evaluation models were asserted to be as important a contribution to clinical prediction as the variables tested by those models. Organization of the Dissertation The first chapter introduced the subject matter of clinical prediction of dangerous behavior as an issue of public policy. The research problem included the need for comparable samples, replicable 19

PAGE 27

methods, and validity for generalizing findings. The theoretical perspectives of civil libertarians and public protectionists provided a background for understanding some practical issues of preventive detention and inaccurate predictions. An overview of the research and methodological difficulties in policy analysis specified the major limitations. Variables reported to be of predictive value were taken from the 1i terature on psychiatric predictions of dangerous behavior reviewed in Chapter 2. The theoretical model was provided by the assumptions and conclusions of previous empirical analyses and the secondary literature generated by policy analysts. Study propositions were derived from the critique of the literature and clinical practices at the public hospital for NGRis studied here. Hypotheses were stated at the end of Chapter 2, and were approved in colloquium in September of 1987. Data sources included hospital and court records and federal and state arrest records. Identification of the sample and descriptive items of criminal and psychiatric history came from the hospital's Forensic Patient Data System, a computerized data base for research and management information. Predictor variables used to test the propositions were listed and defined in Chapter 3, Methods and Design. A description of previously reported probability models was developed to explain notions of false positives and civil libertarian theory. The problem of false negatives was incorporated in the description to explain the public protectionist perspective. The Methods chapter described the calculation of a measure of 20

PAGE 28

the predictive items' Relative Improvement Over Chance (RIOC). The concept represents a slightly different approach in the forensic literature to assessing the value of predictive variables. Discriminant function analysis was chosen to test combinations of predictor variables because of the categorical nature of the outcome groups. Issues of appropriate statistical design for the levels of variable measurement were discussed as one of the data limitations. Selected summary tables reported study findings and statistical analyses in Chapter 4. Various methods of assigning risk level were used retrospectively, and actual rates of dangerous behaviors were analyzed to determine appropriateness of risk assignments made by agency predictors and experimental variables. A model of analyzing false positives, false negatives, and total accuracy rates is in keeping with the previous methods of reporting clinical and statistical predictions. One purpose of additional research was to determine if knowledge about disruptions in patient histories and hospital careers would yield any better accuracy for prediction than chance alone. Gordon's (1977, 1982) alternative analyses with probability theory and his criticisms of assumptions used in most primary data analyses provide a framework for the discussion of findings in Chapter 5. Implications .. for clinicians and their programs for the criminally insane are drawn from the findings. The analyses conclude with recommendations for including events during hospitalization, as well as tradi tiona! fixed variables of age and criminal record, in assigning risk for future danger. 21

PAGE 29

CHAPTER 2 REVIEW OF THE RELEVANT LITERATURE Case law and legislative reform during the study period provided the pertinent legal foundations for understanding the research models reviewed here. United States Supreme Court decisions in 1983 referenced the first-generation core studies of Baxtrom and Dixon patients, and relevant civil case law regarding dangerousness. Legal precedents in criminal law were distinguished by policy analyst Shah (1978) as "judiciary definitions" of danger. The diagnostic model of prediction by psychodynamic formulations (Kozel, 1982; Kozel et al., 1972) provided the pertinent clinical foundations for understanding the theoretical background. Technical approaches to measuring prediction accuracy became more sophisticated in the clinical and correctional literature a generation after the core studies. Monahan (1984) chided the American Civil Liberties Union for concluding a 95 percent error rate for clinical prediction, citing only five studies. Rarely has research been so uncritically accepted and so facilely generalized by both mental health professionals and lawyers as was this first-generation research on the prediction of violence. The careful qualifications the researchers placed on their findings and the circumscribed nature of the situations to which they might apply were forgotten in the rush to frame a bumper-sticker conclusion --'psychiatrists and psychologists can't predict violence'-and paste it on every policy vehicle in sight. (p. 10)

PAGE 30

The review here examined the original core studies to show the tedious development of a body of knowledge from the research. Legal Foundations A review of relevant case law revealed how courts have applied diagnostic concepts to legal definitions and release decisions. The standards of proof (a mere preponderance of evidence, clear and convincing evidence, and evidence beyond a reasonable doubt) serve to allocate the risks of error among 1i tigants. The standards were said by Reisner (1985) to be a function of the severity of conse-quences to society of an erroneous decision. The legal question is one of balancing the social costs of individual liberty against public safety. In Addington v. Texas (1979) the plaintiff argued against a preponderance standard for civil commitment: The individual should not be asked to share equally with society the risk of error when the possible injury to the individual is significantly greater than any possible harm to the state. We conclude that the individual's interest in the outcome of a civil commitment proceeding is of such weight and gravity that due process requires the state to justify confinement by proof more substantial than a mere preponderance of the evidence. (Reisner, 1985, p. 413) Addington's argument that the state should have to prove his need for commitment beyond a reasonable doubt was rejected. The standard of clear and convincing evidence was ruled by the U. S. Supreme Court. The Court rejected the more stringent standard because of the "subtleties and uncertainties in psychiatric diagnoses" (Reisner, p. 414). The ruling was considered to be protectionist and to favor the state's police powers over its protective powers. 23

PAGE 31

Jones v. United States Clear and convincing evidence for the deprivation of liberty was, nonetheless, stronger proof than that afforded the criminally insane in the District of Columbia. The Addington precedent was unsuccessfully used by Michael Jones before the U. S. Supreme Court in his argument against indeterminate confinement as NGRI (1983). Jones contended that his jail sentence, had he been convicted of shoplifting, would have expired after one year; he argued that his hospital confinement should have met the same standard of proof as Addington's clear and convincing evidence for civil commitment. The Court ruled that NGRI acquittees constituted a class of criminal individuals, with due process requirements different from civil standards. The 5-4 decision included references to the literature on psychiatric inability to predict danger by Justice Powell. The public policy problem of the deprivation of liberty was defined in dissent by Justice Brennan, joined by Justices Marshall and Blackmun: The question is whether it is ever constitutional to sentence persons to psychiatric hospitals indefinitely. The question is not whether due process prohibits treating insanity acqui ttees differently from civil committees ... to be balanced: the governmental interest in isolating and treating those who may be mentally ill and dangerous; the difficulty of proving or disproving mental illness and dangerousness in court; and the massive intrusion on individual liberty that involuntary psychiatric hospitalization entails. {103 set. at 3053, 3054) Libertarian Polstein (1985) concluded that "Jones sounded the death knell for the constitutional rights of insanity acquittees" (p. 521). Her critical analyses of the Court's reasoning found little substance to support the majority ruling. 24

PAGE 32

Less analytical in his legal review than Polstein, Singer (1985) found the Court's reasoning problematic, confusing and distressing. He doubted the government would ever again have to prove confin-ability. Margulies (1984) predicted that the Jones decision would shape procedures for insanity commitment and release for years to come. Margulies theorized that commitment really has nothing to do with mental illness or dangerousness, but rather is retaliation: This truncated review failed to take into account the powerful punitive urge which the public and its elected representatives, as well as some of our most noted judges, have displayed toward insanity acquittees. (1984, p. 793) Reform in the direction of more restrictive laws followed the Hinckley (1981) attempt on President Reagan's life. Hearings of the United States Senate were titled "Limiting the insanity defense" (97th Congress). The United States Supreme Court considered psychiatric assessment of danger in their 1983 session, reflecting a public interest in the policy debate. A provocative notion was formulated by Callahan, Mayer, and Steadman (1987). Their theory led to the study of legislative changes made in the three years prior to the U. S. Supreme Court ruling in Jones and the three years after. The Steadman group suggested that the Hinckley trial and NGRI verdict of 1982 was not the moving force behind increased conservatism. They posited, rather, that Jones legitimized current practices of prolonged and indefinite hospital stays. Some evidence of legislative response to the Supreme Court ruling was offered, but the events occurred in such temporal proximity (Hinckley, Jones, legislation) that their findings were inconclusive. 25

PAGE 33

Less offended by the Jones holding than Singer or Margulies, Hermann (1983) picked up on a footnote of procedural protection, that Jones had voluntarily entered an insanity plea to shoplifting. When the defendant himself raises the defense, automatic commit ment may be more justified. Hermann implies the very culpability the NGRI was relieved from (that is, an absence of mens rea appropriate to the elements of the offense). Similar to this logic is the required competency in Colorado on the part of the defendant before entering a plea of insanity. Hermann supported some of Justice Brennan's dissent, with assurance that states with less restrictive laws than the District would not be required to conform with the more restrictive federal ruling about automatic and indeterminate confinement. Schmidt (1984) lamented that Justice Powell's "common sense" judgments flew in the face of social science research. The majority wrote "It comports with common sense that someone whose mental illness was sufficient to lead him to commit a criminal act is likely to remain ill and in need of treatment" (3050). Schmidt found the premise untenable in law or in social science. Legal reviews of the policy implications of Jones tended to be libertarian, and several suggested that presumption of dangerousness and continued insanity simply assures confinement. Legal Definitions of Danger One policy analyst defined the prediction problem as a broader, conceptual issue that went beyond law and psychology (Shah, 1978). In an article suggesting a research paradigm that would distinguish 26

PAGE 34

between dangerousness and violence, and between legal definitions and clinical, Shah reviewed "judiciary" definitions. A federal judge saw danger in a defendant's "check writing proclivities" in Overhol-ser v. Russell (1960). In an indecent exposure case, Carras v. District of Columbia (1962), the court reasoned that sexual danger was not restricted to physical injury alone. The Carras court wrote of potential harm including psychological harm to "sensitive adult women and small children" (1978, p. 226). A few years later in Millard .v. Harris (1968) the same federal jurisdiction allowed that only a small minority of "supersensitive women and small children are likely to suffer serious harm from isolated instances of exhibitionism" (1978, p. 226) and released Millard. The U. S. Court of Appeals noted that "a finding of danger-ousness must be based on a high pro ba bili ty of substantial injury" in Cross v. Harris (1969). Judge David Bazelon cautioned against labeling as dangerous anyone we preferred not to encounter, and assumed that Congress had not intended dangerous to be used "in any such Pickwickian sense." (1978, p. 226) Shah's final example was from the Supreme Court of New Jersey in State v. Krol (1975) and is relevant to outcome definitions with the Colorado sample. Dangerous conduct is not identical with criminal conduct. Dangerous conduct involves not merely violation of social norms enforced by criminal sanctions, but significant physical or psychological injury to persons or substantial destruction to property. Persons are not to be indefinitely incarcerated because they present a risk of future conduct which is merely socially undesirable. (1978, p. 226) Colorado's district courts continue to debate whether an alcoholic's self destructive and socially disruptive behavior constitutes an 27

PAGE 35

abnormal mental condition likely to cause danger to society. In Stephens v. Colorado (1987) the patient claimed that he had feigned insanity in 1977, and that his alcohol problem did not constitute an abnormal mental condition that caused him to be dangerous. The prosecution, in a contested conditional release hearing, argued that Stephens had been institutionalized (incarcerated for 30 of his 52 years), couldn't fend for himself, and might drink himself to death or get killed on the railroad tracks in Chicago. A psychiatrist testified that the patient was not treatable, did not have a "disease or defect" and that society could not provide "brother's keepers" for all such persons. The judge noted that "abnormal mental condition" included personality disorders as far as he could tell. The judge acknowledged the clinician's aversion to warehousing people, but concluded "This statute is primarily designed for the benefit of protection of the people" (p. 9). The patient was denied release for his long, disruptive, and essentially nonviolent criminal career. Outcome Criteria for Danger First generation research counted arrests, usually making no distinction among types of criminal charges. Rehospitalization had been defined by Baxtrom and Dixon researchers as "recidivism". The Maryland group in 1984 were among the first to report the legal disposition that followed disruptive or criminal behavior. If one only counts a subsequent NGRI finding as "recidivism," then extremely low rates are reported. In Colorado (1977) and for the decade since, rates of new NGRI commitment were at one and two percent. 28

PAGE 36

Conviction and correctional incarceration rates among former mental patients appear much higher than recommitment as NGRI, confirming that a history of legal insanity was no guarantee of a successful plea at a later time to new criminal charges. Mental health diversion was reported to be common, but assaultiveness was the only behavior defined as dangerous. Rubin (1972) defined the assaultive crimes of homicide, assault, armed robbery, and rape as violent crimes. Shah (1978) agreed with that as an operational definition of "violent crimes." Violent behavior on condi tiona! release is not required, however, for initiation of revocation procedures in Colorado. The statutes allow the patient to be declared no longer eligible to remain on the status for any violation or criminal behavior, as a result of an abnormal mental condition. That is, the danger criterion is invoked again in the original court of commitment. Because of the official nature of revocation, that outcome was defined as a dangerous disruption in this research. One premise underlying the design of this study and the outcome variables for measuring disruption after conditional release was an assumption that relapse for a chronically schizophrenic mental patient is not synonomous with dangerous behavior. When relapse occurred, in the absence of crimes of violence or revocation, disruption was classified as nondangerous disruption. Rehospi talization through voluntary or civil action was to be expected, and may have served to prevent further decompensation and dangerous behavior. 29

PAGE 37

Psychological Assessment of Danger The use of psychological tests for more valid and reliable measurement of a propensity for danger was described as one diagnostic approach reported in the literature. Public protectionists who criticized the early conclusions (overclassification of danger and the generalizability of findings) were studied for alternative inter-pretations of the research problems. Floud and Young (1982) philosophized preventive confinement for those whose dangerous behavior was "willful harm" and contended that there is no way to avoid the concept of guilt and just dessert. The British policy study cautioned, however, against detaining persons for what they might do in the future on the basis of their thoughts and ideas. A psychological trait, like aggressiveness or hostility, was defined as a relatively stable personality characteristic that might be perceived as dangerousness by the evaluator. Personality traits were clinically related to the idea of a person's propensity to injure or harm another. Quite apart from social situations or environmental circumstances, dangerousness was hypothesized as a personality characteristic, much like intelligence or aptitude. The concept implied that some driving, destructive force resides within the dangerous offender on a permanent basis. Greenland (1985) indicts psychiatrists for assuming a notion of individual violent tendencies: Although most physicians recognize the close relationship between poverty, powerlessness, and disease, the impact of this association on the practice of medicine has been very modest. This is especially so in forensic psychiatry, which still acts as if the impulse to rape, maim, or kill resides like malignancy within the individual offender.(p.38) 30

PAGE 38

Even if such a malignancy could be identified, of greater significance to prediction accuracy is the presence of similar impulses among safe individuals. Hostile, aggressive, destructive urges and ideas may be common to human experience, and only rarely lead to violent acts. Personality Scales Validation studies on trait theory were attempted with personality tests, and similar psychometric approaches, to try to distinguish dangerous from safe persons. Megargee (1976, 1981) postulated that certain items on the MMPI (Minnesota Multiphasic Personality Inventory) could measure overand. undercontrolled hostility, each extreme considered as a propensity for dangerous behavior. A new scale (OH) score was obtained-by factor analysis. The overcontrolled person tended to be passive, prosocial to the point of rigid values, nonexpressive of emotion, and uncomfortable with his own and others' aggression. Such a person could behave in violent ways when controls broke down and released pent up hostility. The undercontrolled person was thought to be impulsive, overtly angry, narcissistic, and antisocial in values. His hostility was not pent up, but was released by his lack of controls. The underlying construct of hostility was thought to be a proxy measure for the concept of a trait of dangerousness, with both the overcontrolled and undercontrolled at high risk. Megargee's work was methodologically sound, with replicable items and scoring techniques. The problem of unacceptably high 31

PAGE 39

rates of false positives, that is safe persons who tested dangerous on the OH scale, was not overcome in subsequent tests of validity with the MMPI on a number of criminal, mentally ill, and normal populations. After more than a decade of refinement, Megargee's 1970 observation was confirmed. "No methods have been devised which will adequately postdict, let alone predict, violent behavior" (1970, p. 145). The finding was based on his inability to distinguish prison inmates with violent crimes from inmates with nonviolent crimes, or prison from non prison samples, some of his tested criterion variables for danger. Psychological trait theory had a major influence on attempts to identify a characteristic of dangerousness or a propensity for violence within the offender. Clinical tradition in personality assessment favored a projective technique of the Rorschach inkblots as a diagnostic test of angry, hostile, and aggressive ideas and urges. There is little substantive evidence of Rorschach validity, however, in the law or the literature on prediction of danger (Webster et al., 1985). Meloy (1988) a psychodynamic practitioner who works with offenders in southern California, postulated two types of violent persons and disagrees with the majority critics of Ute Rorschach. Meloy classifies dangerous persons as "affective" or emotional, or as "predatory", and reports a number of different profiles on standard psychological tests (1990). He places emphasis on early childhood bonding experiences that influence motivations of fear and anger later on in life. Unlike first generation researchers who found 32

PAGE 40

diagnosis not very helpful, Meloy and other recent reports have resurrected Cleckley's (1964) psychopathic profile and associated the diagnosis with violence. The Bridgewater legacy and recent attempts at sex offender typology fit a similar philosophical model. Contemporary studies of violence tend to focus on short-term classifications for management of acute psychiatric patients in hospital settings. Yesavage (1984) measured schizophrenic symptoms on the Brief Psychiatric Rating Scale (BPRS), serum level of a standard dose of neuroleptic drug, social and psychiatric history, and disruptions during the first eight days of hospitalization. This Veterans Adminstration hospital found that Vietnam combat experience was correlated with in-hospital disruption. Klassen and O'Connor (1988) suggest that we may never achieve improved accuracy rates for longer range prediction, but they chose a mid range follow-up of about six months. Theirs is a refreshing focus on situational variables, albiet not very practical ones. For example, the amount of time since last intercourse, injuries experienced at the hands of a sibling before age 15, and the like. Family, friendship, and work environments were evaluated for predictive value. With an N of 239 adult males they used discriminant analysis with 67 potential predictor variables. Parent loss before age 15, self-reported violence, and dissatisfaction with family were associated with subsequent violence. The Bridgewater Legacy One of the first published attempts to evaluate the diagnosis and treatment of dangerous sexual psychopaths came from the 33

PAGE 41

security hospital at Bridgewater, Massachusetts (Kozel, Boucher, and Garofalo, 1972). Evaluators formulated psychodynamic concepts of personality and behavior thought to be related to characteristics of dangerousness. The multidisciplinary team reported solid interrater reliability, suggesting sophistication in operationalizing their definitions. Similar concensus with a rating instrument (TRIAD) used in a California psychiatric emergency room (Segal et al., 1986) also reflected a common philosophical base with good reliability among raters of short term danger. Validity of longer term danger assessments, and validity of the psychiatric nomenclature, is an issue separate from reliability among raters. Whether either group actually measured dangerousness, rather than some social value, attitude, biochemical state, or psychiatric symptom, could be known only through replicable measures. The interrater reliability suggested common decision rules within each setting, but did not establish construct validity. Psychopathy and Validity of Diagnosis A seminal work by libertarian and activist Bruce Ennis (Ennis & Litwack, 1974) was a classic commentary on the imprecise nature of psychology and psychiatry. Ennis showed how nonscientific and arbitrary psychiatric testimony was. Behavioral criteria for diagnostic nomenclature improved with the revised, third edition of the Diagnostic and Statistical Manual of the American Psychiatric Association (DSM III, 1985), but the personality disorders remained an enigma for practitioners and researchers alike. The diagnosis of psychopath, sociopath, and antisocial personality disorder, has been referenced in 34

PAGE 42

and out of statutory definitions and diagnostic nomenclature as long as insanity law has been recorded. Clinicians believe it is easy enough to diagnose, but policy makers disagree on whether it is a form of mental illness or insanity. After a career devoted to public sector forensic psychiatry, Kozel (1982) offered the following definition of the dangerous person, which sounds much like diagnostic descriptions of psychopathy: One who has actually inflicted or attempted to inflict serious physical injury on another person; harbors anger, hostility and resentment; enjoys witnessing or inflicting suffering; lacks altruistic and compassionate concern for others; sees himself as a victim rather than as an aggressor; resents or rejects authority; is primarily concerned with his own satisfaction and with the relief of his own discomfort; ... has immature attitudes toward social responsibility and distorts his perception of reality in accordance with his own wishes and needs. (p. 251) The methodological problem is finding valid measures for identifying just which offenders, in Kozel's words, still "harbor anger" or "see themselves as victims." The concepts are clinical and interpretive, rather than behavioral, making operational definition difficult. The Meloy work, and the British Columbia work (Hart et al., 1988) with the Hare Psychopathy Checklist, reflect a psychodynamic philosophy similar to the Kozel group. More recent work from the same Massachusetts facility (Rosenberg et al., 1988) did report, as did the Canadian group, more technically sophisticated methodological approaches. Their predictor variables were quite clearly based on a similar, nonsta.tistical classification theory reported by an earlier Bridgewater group (Cohen, Garofalo, and Boucher, 1971). Impulsivity and rage are examples of motivating factors that differ among sexually ass a u1 ti ve persons in their typology. 35

PAGE 43

Prediction Accuracy The Bridgewater population was probably the largest and most dangerous group (n=435) of psychiatrically treated sex offenders ever studied (Kozel, 1982). The Massachusetts treatment emphasis was to help offenders become prosocial and morally responsible. Recommendations were made to the courts about risk level at the time of release, with recidivism (from arrest records) used as a method of measuring accuracy of the danger classifications. Offenders predicted to be dangerous did, in fact, have higher recidivism rates (35 percent of 49 persons evaluated dangerous) than those predicted to be safe (8 percent of 82 persons). The Kozel et al. (1972) study is referenced as one of the few with primary data, and is used by most policy analysts to show over prediction of dangerousness, with two out of three high risk offenders not displaying dangerous outcomes. Critics complained that the followup time was too variable; so the Kozel group continued work with the sample for about 10 years, confirming similar rates among an accumulated 592 patient evaluations. Gordon (1977) compared the Massachusetts' relatively high (.35) base rate, however, to his study of Patuxent patients (.41) and a Colorado sample (.48), the proportion of high risk patients with dangerous outcomes. Did those samples (Massachusetts, Maryland, and Colorado) provide enough evidence of external validity to warrant the generalizations of unacceptably high rates of false positives made in the secondary 1i terature? Gordon and this researcher think not. Three reported probabilities were near even chance, a great deal higher 36

PAGE 44

than the general population's chances of behaving dangerously. Considering our clinical ability to recognize safety in a general population, Gordon contended that to conclude that we cannot predict danger is also to conclude that we cannot predict safety, when in fact, "Most of us are predictably quite safe." (Gordon, 1982, p. 296) Replication of the group's findings was precluded by lack of sufficient criteria or decision rules to substantiate their reported reliability among raters of danger. The distribution and homogeneity within a sample group of offenders drives this and any other study lacking cross validation. The kind of patients at Bridgewater excluded psychotics and included personality disorders with sexually assaultive crimes. Psychiatric hospitals are not likely to have such samples for study, at least not from the present decade. Elimination of the irresistible impulse from the Colorado insanity test (a volitional test that said the defendant knew right from wrong, but could not choose the right and refrain from the wrong) eliminated most defendants with primary personality disorders (PDS reports, 1981-1989). The present study included less than a dozen who would fit the description of Bridgewater subjects. The Baxstrom and Dixon Cases A ruling by the United States Supreme Court in Baxstrom .Y:. Herold (1966) affected nearly 1,000 mental patients housed in New York security hospitals. The Court held that inmates who became mentally ill in prisons had rights to due process at the time of 37

PAGE 45

transfer to state hospitals for the criminally insane at Dannemora and Mattawan. Inmates were entitled to be transferred to civil hospitals when their sentences expired. Most were manageable in civil hospitals and many died there. The patients affected were the oldest in the New York facilities, with almost 40 percent over the age of 70 at the time of the court ruling (Steadman and Cocozza, 1974). Researchers found only three studies of the criminally insane in the international literature at the time of the Baxstrom followup in 1970, and one of those was an unpublished study at Colorado State Hospital (1966). The early studies were the first demographic descriptions of persons broadly defined as criminally insane. The mean age of Baxstrom cases was 51.6 years and 45 years for the Colorado sample. New York cases had an average length of hospital stay of 14 years, compared to the Colorado report of an average length of stay of 2.5 years for NGRis released in 1962-64. Monahan (1981) called the long term detention of Baxstrom patients a result of bureaucratic inertia, and Greenland (1985) said the studies provided ... depressing information about slovenly administrative and diagnostic procedures." (p. 31) Criterion validity was based on only 98 (of 969 patients in the Baxstrom class) who were eventually released to community placements and followed for rearrest or rehospitalization. The 98 Baxstrom releases were followed after an average of four years at risk in communi ties. Rearrest for index crimes of homicide, rape, aggravated assault, and armed robbery had been previously used definitions of Rubin (1972) as both determinants, that is, predictors, and as the 38

PAGE 46

outcome or criterion variables. New York researchers learned that not all assaultive behavior resulted in arrest or criminal charges. Diversion of former mental patients into mental hospitals was thus included as dangerous behavior when assaults occurred. The Legal Dangerousness Scale The Baxstrom researchers explored dozens of variables kept in institutional records and New York's central Department of Mental Hygiene, and tested them for predictive association with dangerous outcomes. Assaultive behavior during hospital career, psychiatric diagnosis, and hospital length of stay were examples of variables found not to discriminate the dangerous from the safe. The patients were old and institutionalized, so history was often remote. By restrospective statistical analyses of social and psychiatric events in patient histories, Steadman and the New York group concluded with a four item Legal Dangerousness Scale (LDS). They assigned a binary coding score for each i tern of criminal record, resulting in a perfectly reproducible summary measure. Reproducibility came from the fact that only one combination could yield each score from 0 to 15. Points were summed, and a cutting score of 5 and age under 50 designated the high risk group (31 percent of 36 patients were subsequently dangerous), with the low risk group having only 5 percent dangerous outcomes among the 62 patients predicted to be safe by the LDS and age risk assignment. The scale was a relatively simple and straightforward approach that objectively assigned offenders either to a high or low risk group, according to events of criminal history and age. The New 39

PAGE 47

York reports ascribed Guttman type properties of ordinality to the items, and assigned unequal weights (actually a binary code) to four categorical events of record. A coefficient of reproducibility of .91 was reported as evidence that the items were scaleable. The scoring technique that implied that juvenile record (8 points) was twice as important as previous incarceration (4 points), and constituted automatic assignment to the high risk group, was questioned in a validation study of the LDS properties by Koppin (1977). Reported school disruption, previous psychiatric hospitalization, and nonviolent crimes were also associated with subsequent dangerous behavior. The youngest age group had proportionately less dangerous behavior than men in their thirties and forties. Ordinality, as least in the Guttman sense, was not confirmed in the validation study with an N of 111 and a coefficient of reproducibility of .83. The published Baxstrom reports, including the Cocozza dissertation of 1974, did not describe the research method used for derivation of weights, and unequal weights were not confirmed in the replication work. The scoring system did have value simply as a code for identifying which variables were present in the offender's history. There was only one possible combination of LDS items for each score: Juvenile record = 8 Previous incarceration = 4 Prior violent crime = 2 Severe instant offense = 1 Guttman properties suggested that if the first item were present, all remaining i terns would also be present. That was not confirmed, but 40

PAGE 48

predictive value of the LDS variables was found with the Colorado sample. The LDS high risk group of 60 had 48 percent dangerous behavior after release, with only 8 percent of 51 low risk patients being dangerous. Base rates of total samples were remarkably similar, but various combinations of criminal and social histories yielded about the same predictive value as the New York scale items. At best, prediction of danger was accurate in about half of the cases of high risk, and total accuracy rates hovered near the base rate of the criterion behavior. There were few empirical investigations, and even fewer techniques for comparing samples, methods, or findings, when the Legal Dangerousness Scale was published in 1974. Policy analyst Gordon used the scale as a comparative statistic in public policy studies of Maryland's defective delinquents in 1977 and 1982. Correctional literature has continued to use criminal history as one of the best discriminators for identifying parole success and criminal recidivism. Political Predictions and Dixon Cases -----Strikingly similar to Baxstrom findings, mass transfers of patients from Pennsylvania security hospitals to civil settings followed a court decision in Dixon v. Attorney General of the Commonwealth of Pennsylvania (1971). These 586 patients were also not specifically predicted to be dangerous, although 114 were later opposed for transfer on those grounds. All Dixon patients had been held in secure hospitals for an average of 14 years, without appropriate due process under the Fourteenth Amendment. 41

PAGE 49

Researchers broadly defined prison transfers as criminal commit ments, along with those under a defective delinquent law for indeter-minate sentences. The affected Dixon class had become civil recommitments after expiration of criminal statuses, without equal protection. Fourteen percent (of 438 patients) were considered dangerous after a four year follow up by Thornberry and Jacoby (1979). Like Baxstrom cases the mean age was 47 at the time of transfer and many had died by the time of follow-up. Unique in the literature, the Pennsylvania researchers followed their patients and asked them about their hospital experiences, current living situations, employment, finances, and general well being in community placements. A replicable aspect was their use of the Katz Adjustment Scale (Katz and Lyerly, 1963) for clinical adjustment and level of functioning measures. The instrument had also been normed on nonpsychiatric populations, and had baseline data on "typical" mental patients. Dixon patients did not score well on these indices, but were uniformly happier being out of the security hospital than in. Prediction Accuracy Thornberry and Jacoby (1979) retrospectively applied the New York Legal Dangerousness Score, with less discrimination found for identifying the dangerous among the Dixon class. Several years passed before the results were published amid some skepticism surrounding other right-to-treatment cases in Pennsylvania. Their study is the only one known to have included "self-releases" by escape (p. 193). Twenty-four percent of all releases were arrested at 42

PAGE 50

least once, and 14 percent were identified as dangerous, defined as behavior injurious to another person. Time at risk was as little as less than one month to over four and a half years. Pennsylvania researchers also used assaultive behavior in the civil hospitals as an outcome criterion, and found that the security hospital's assessments of danger were exaggerated. The authors concluded that the decision making processes were not actually clinical, nor statistical, but rather political and concerned with political consequences: We would submit, however, that there is a third type of predictive exercise ... which we will call political prediction. It is based, not on the characteristics of the individual, but on.the assumed characteristics of a group to which the individual belongs. In essence, group predictions are projected onto the members. When the process is properly conducted the probability of group events, and the deduced individual event, is based on grouping subjects together on the basis of variables known to be associated with offending. In political prediction, groups are created on the basis of criteria which are assumed, but not demonstrated, to be correlated with offending (Thornberry & Jacoby, 1979, p. 27). Examples of such groups are the mentally ill, felons, murderers or rapists, psychopaths, schizophrenics, or perhaps in the present study, NGRis. Research Propositions Few studies with precisely defined samples according to NGRI legal status were reviewed here. Specific statements about a patient's likelihood to be dangerous, as a result of an abnormal mental condition, were not operationalized in the clinical or legal literature. Clinical predictions were either inferred retrospectively 43

PAGE 51

or were vaguely related to unspecified and nonreplicable treatment programs. Nearly all were based on zero-one probability models of dichotomous prediction and outcome groups. Assignment to risk groups on the basis of statistical frequencies typically yielded better rates of predictive accuracy than clinical methods. It was assumed that methodological approaches to evaluating accuracy of danger classifications were as relevant to rational policy debate as predictor items. The following research statements were formulated to test methods and statistical techniques, as well as to test for variables to assign patients to high or low risk groups. Models of actual and equal probability, and measures of improvement over random distribution (chance) were the statistical approaches to the research questions identified in the review of the literature. Proposition .! Agency opposition to release constitutes a prediction of danger and will distinguish the dangerous from the safe after release. Proposition l Groups defined by agency recommendation for or against release will themselves be distinguishable by disruptive events in patient histories. Proposition 3. Disruptive events in precommitment histories and hospital careers will distinguish the dangerous from the safe after release. Age, sex, race, criminal record, and certain in-hospital events were tested for predictive value in identifying the dangerous after release. Statistical models are described in the next chapter. 44

PAGE 52

CHAPTER 3 METHODS The study was a retrospective analysis of the outcomes of 109 NGRI patients who obtained conditional releases trom the forensic unit of Colorado State Hospital between 1981 and 1986. Hospital recommendations to the criminal courts of commitment constituted a clinical prediction of each patient's likelihood of future danger, as a result of an abnormal mental condition. NGRis who were released by the courts against hospital recommendation (n=31) were hypothesized to be at higher risk for danger than NGRis whose release was recommended (n=78). The release cohort was followed after a minimum of two years at risk in the community, for psychiatric relapse and criminal recidivism. Research groups were thus defined by clinical prediction and by type of outcome. Revocation of conditional release and/or subsequent crimes of violence were defined as dangerous disruption after release. Nondangerous disruption included rehospitalization or rearrest for nonassaultive behaviors or relapse. The design, including the predic-tion models replicated from previous research, examined some theoretical and empirical relationships among pre-hospital, in-hospital, and post-hospital disruptions. Statistical relationships among disruptive histories and release outcomes were explored to find methods to discriminate dangerous from safe patients. Details of the research process are described under the following headings:

PAGE 53

1. Design and Prediction Models 2. Selection of the Sample 3. The Treatment Setting 4. Description of the Sample 5. Release Procedures 6. Data Sources and Limitations 7. Operational Definitions 8. Statistics Design and Prediction Models Like the Steadman-Morrissey (1981) study of New York incompetent to proceed cases, the data analyses were retrospective, but predictor variables were applied as if they constituted prospective scales for high and low risk. The Steadman-Morrissey approach tests the proposition that actuarial approaches would have been as accurate as clinical predictions, and that either would have been more accurate than chance. Variable reduction was achieved with discriminant function analyses of predictor i terns to distinguish dangerous from safe outcomes as a dependent variable. Similar analyses of the i terns were applied to the dependent variable of hospital opposition (that is, the prediction of danger). Steadman and Morrissey (1981) compared their design to a tradi tiona! cross validational model. A Bayesian (actual) probability model consistent with a civil libertarian perspective, and an equal probability model consistent with a public protectionist perspective, were applied to determine the better accuracy and the social costs of one model over the other. 46

PAGE 54

The libertarian model emphasized the goal of reducing false positives; the protectionist model emphasized the minimization of false negatives. Predictions, or assignment of high or low risk for future danger, were treated as independent variables. The dependent variable, dangerous behavior after release, was deduced from disrup-tive outcomes, including civil and voluntary rehospitalizations. A two-by-two contingency table shows the prediction (independent) variable in the columns and the outcome (dependent) variable in the rows. Outcome Safe Risk Assignment Groups Safe Prediction Dangerous True Negative A False Positive B Dangerous False Negative c True Positive D Figure 1. Prediction by Outcome Contingency Table. A. True Negatives: persons (accurately) predicted to be safe who were, in fact, safe after release. B. False Positives: persons (inaccurately) predicted to be dangerous, who, in fact, were safe after release. 47

PAGE 55

C. False Negatives: persons (inaccurately) predicted to be safe who were, in fact, dangerous after release. D. True Positives: persons (accurately) predicted to be dangerous who were, in fact, dangerous after release. Additional analyses involved a similar, but two-by-three, contingency table with three outcome groups, in order to capture disruptions in the community that were not associated with violent crime or revocation. The dependent variable in those tables thus became three rows for NO DISRUPTION, NONDANGEROUS DISRUPTION, and DANGEROUS DISRUPTION. Selection of the Sample A group of 144 NGRis who obtained their first release from criminal court commitment after June 30, 1981 and before July 1, 1986, were identified through the Patient Data System (PDS) micro-computer search. Nineteen of 24 uncondi tiona! discharges were NGRis released to law enforcement detainers to face trial or serve sentences on prior criminal convictions. Direct discharges consequently had less opportunity for criminal recidivism in the community than condi tiona! releases. The researcher also had less information on which to evaluate outcome. Discharge by death {8) and unconditional discharge (24) were thus excluded from the follow up sample, as were (3) condi tiona! releases lost to follow-up. A sample of 109 males conditionally released for the first time from NGRI commitment was selected. Included were patients with sporadic contact with the hospital, mental health centers, families, 48

PAGE 56

police, or fellow patients. The majority remained on conditional release status for at least one year, with the aftercare staff providing written progress reports to their courts of NGRI commitment. Outcomes from all sources were included in the data when validated by law enforcement or hospital records. A small group of continuously enrolled re-releases, recycled on former court orders of release, were excluded as atypical of the release process. Wide variation in their lengths of initial stay, and subsequent rehospitalizations ranging from days to years, made calculating time at risk awkward and cumbersome. Also excluded were two cases whose conditions of release included "signing in voluntarily" for continued forensic hospitalization. NGRI women were excluded from the sample because they were in nonforensic programs. All remaining conditional releases were identified from PDS listings of status and enrollment changes between June 30, 1980, and July 1, 1986. The final sample included 109 first conditional releases from NGRI commitments, placed in communities rather than other correctional institutions, with opportunity for psychiatric relapse or criminal recidivism. Diagnostic and demographic characteristics are described in Description of the Sample. The Treatment Setting The Colorado State Hospital is fully accredited by the Joint Commission on Accreditation of Healthcare Organizations, a private professional standards agency of the medical professions. Hospitals 49

PAGE 57

voluntarily subscribe for rigorous surveys and evaluation for compliance with JCAHO standards. The hospital is state licensed as a 700 bed public mental hospital in Pueblo, Colorado. At the time of study, the hospital was organized into clinical divisions, the largest one a 350 bed specialty unit for forensic psychiatry. This division was named the Institute for Forensic Psychiatry (IFP) and, at the time of the inpatient careers of this NGRI sample, was organized around three specialty programs located in five buildings. Treatment Tracks I and II had physical facilities at each of four security levels. High security levels were maximum and medium units with fences, electronic surveillance, and guards. Low security levels were intermediate and minimum units, similar to nonforensic adult psychiatric wards. Ward level care was given by registered nurses, mental health workers, and licensed psychiatric technicians. Female staff were involved in clinical treatment on all security levels. The male to female staff ratio was about 70/30 during the study period. Program Description Track I was the larger treatment program with an average daily NGRI census of 120, compared to 90 NGRis on Track II. Each program also treated incompetent to proceed to trial and civil patients, and Track II had several convicted sex offenders under parole board transfer. In any typical daily census, NGRis comprised about two-thirds of the total number of IFP patients. The Track I program was for chronically mentally ill persons, often psychotic and socially inadequate, or debilitated by symptoms of institutionalization. 50

PAGE 58

Supportive, individualized programs for daily living and basic communication were formulated by core professional staff for continuity of care along security levels. Antipsychotic drugs were commonly prescribed to bring patients into sufficient remission for home groups, problem solving, and reality therapy. Didactic material on major mental disorders was routinely included in small group therapies. Psychotropic medication was less routinely used on Track II because of the diagnostic case mix, although about a third of the Track II patients were well compensated from acute psychoses. While many Track II patients had long, antisocial histories, they were more socially outgoing and verbal than those classified as Track I. Track II patients often carried secondary diagnoses of personality disorders or substance abuse. A therapeutic community model of patient government, mutual responsibility, peer confrontation, and insight oriented psychotherapy was followed across Track II security levels. Self help groups, didactic programs, and psychological literature were available to Track II patients from medium security forward. As patients developed their own internal controls and complied with treatment expectations, they were rewarded with increased privileges and transferred to lower levels of external security. Gradual increments in the privilege process were common to both treatment programs, with incentives for cooperation in treatment. Both programs transferred patients to higher levels of security when believed necessary for external control and management of illness or behavior. Behavior management transfers (BMTs) were defined as in-hospital disruptions, along with escapes, and leaves for jail or 51

PAGE 59

court appearances. Track I patients spent proportionately more time on low security areas and Track II patients spent proportionately more time on high security areas. Prerelease Testing in the Commrmi ty Permission from the court of commitment was required for any treatment or rehabilitation activities off of the hospital grounds. District attorneys had an opportrmity to contest the granting of a court order sanctioning such activities, and arguments were subsequently heard by the judge. Progression to rmsupervised passes in the community was implied by clinical and administrative reviewers as a necessary release criterion for patient release eligibility. Passes were used for work, school, home visits, or approved recreation in the local commrmi ty. Passes were generally considered to be a time for gradual re-entry and testing of the patient's internal controls. Patients were expected to bring back to therapy the experiences and accomplishments from their time away from the structure of the hospital. Abstinence from street drugs and alcohol, and discretion in sexual relationships were encouraged for patients in both programs. Day passes were given in the beginning, followed by overnight, weekend, and a week or more. Supportive or independent 1i ving arrangements were used in the commrmity, according to the patient's level of frmctioning. 52

PAGE 60

Maximum Security Programs The third program, sometimes called Track III but located in only one physical facility, was the program of Special Services, a maximum security unit for all IFP admissions. Acute care and shortterm evaluations were housed on a four-ward, 80-bed security hospital built in 1974. A yearly intake of some 600 patients included civil cases from county jails, criminal court pretrial evaluations, correctional transfers, incompetent to proceed to trial cases, and civil commitments considered unmanageable by the civil hospitals. About a dozen highly dangerous or disruptive NGRis were in maximum security treatment programs, and a few of those were discharged to correctional detainers during the time period studied here. They all had been tracked and treated in less secure program assignments, and had then been transferred (by BMT) to maximum security prior to release. Description of the Sample Forty-four NGRis were conditionally released from treatment Track I, with 7 official agency oppositions to their release. Sixty five patients were released from treatment Track II, with 24 releases opposed by the agency but approved for release by the courts. Most patients (64 percent) were released from low security treatment areas, with the hospital opposition rate higher for those seeking release from high security areas. The sample is representative of only a small proportion of NGRis treated during the years of study. In one sense, the sample is 53

PAGE 61

a "universe" of all first conditional releases within a certain time frame. Their commitment origins spanned many decades, however. The better method for sampling typical practices would have been to take an admission cohort. Greater generalizability could have been achieved with patients admitted under similar policies and programs, rather than release cohorts like these samples. Psychiatric diagnoses given by psychiatrists at the time of examination for release confirmed significant program differences. Primary psychoses were diagnosed for 84 percent of Track I and 43 percent of Track II patients. Substance abuse was primary for less than 5 percent of Track I, compared to 29 percent of Track II patients. Other nonpsychotic disorders were more prevalent on II, and in general, patients with personality disorders were more often opposed by the hospital than those with major psychotic illnesses. Demographic Characteristics The Denver metropolitan counties accounted for 56 of 109 commitments. Age at the time of release ranged from 19 to 64 years, with an average age of 34. Ethnic minorities included 20 percent Hispanic and 15 percent Black patients. Crimes of violence (homicide, sexual assault, aggravated assault, and armed robbery) were involved in 61 percent of the NGRI commitments, with Track II having proportionately more crimes of violence (71 percent) than Track I (48 percent). Nearly half of the total sample had never been married, a high risk indicator reported by Klassen and O'Conner (1988). Poor job histories were reported for patients from both treatment tracks. Track II patients had more high school graduates and persons who 54

PAGE 62

had attended college (a cumulative 60 percent). Their level of social functioning was one rationale for the cognitive program approach. Length of Hospital Stay and Crime The average length of stay was 5.4 years, ranging from the shortest of 3 months to the longest of 17 years. The extremes in length of stay were infrequent, with the median at 55 months, or just under 5 years. Previous NGRI release studies from this hospital reported 2.5 years in 1966 and 3.7 years in 1977. Limitation on any generalization is reflected by the small sample size, and the frequencies appearing in the categories used to describe the most serious crime of commitment. Average lengths of stay by crime category appear in Table 3.1. Lengths of stay less than one year were more common in the early years when there were 12 short stays. Table 3.1. Crime of NGRI Commitment by Length of Hospital Stay in Months: Conditional Releases 1980-1986. (n=109) Crime n Min. Max. Mean S.D. Homicide 12 23 199 95.33 53.06 Sexual Assault 10 34 184 81.20 47.79 Aggravated Assault 32 10 185 69.00 43.37 Kidnapping 2 20 36 28.00 11.31 Armed Robbery 10 16 178 66.40 53.17 Simple Assault 8 5 76 39.38 25.01 Arson 5 3 122 39.60 47.69 Criminal Mischief 2 9 59 34.00 35.36 Burglary 14 7 134 60.57 41.82 Trespass/Theft 11 5 203 54.18 54.23 Forgery/Fraud 2 26 41 34.50 9.19 Nonassaultive Sex 1 83.00 Total 109 3 203 64.78 46.49 55

PAGE 63

A total of 44 patients accounted for 100 escapes, usually walkaways from privileges. The average length of time on escape was 247 days, skewed by a few escapes of several years' duration. Length of hospital stay included time absent on escape, pass, or leaves to court. In summary, length of stay was not a straightforward measure. Release Procedures The interface between legal and clinical mandates to protect public safety and provide psychiatric treatment and rehabilitation, requires complex tradeoffs between social costs and benefits. Decisions to detain some NGRis and to conditionally release others was considered a risky business by clinicians and courts. Implied tradeoffs were retrospectively evaluated for this study from libertarian and protectionist perspectives, after actual decisions were made. The study of legal and clinical criteria that may have been used by decision makers helped define the research problem. One consideration for psychiatrists and lawyers was the difference between release criteria and commitment criteria. The Legal Release Criteria The proof of insanity at the time of the initial trial did not require the proof of dangerousness, nor proof of the necessity of hospitalization. Although NGRI commitment law required no presumption of future danger for initial aquittal or automatic hospitalization, once adjudicated in Colorado, the law assumed an abnormal mental condition caused the NGRI patient to be dangerous and in need of 56

PAGE 64

hospital confinement. Temporary insanity was not recognized in the Colorado statutes, and at the time of study a minimum of six months in the hospital was required before the patient could file a motion for a release examination and hearing. In addition to the presumption of a continuing abnormal mental condition, Colorado NGRis were legally presumed to be dangerous because of the release eligibility law, not the insanity commitment law. A broad definition of future dangerousness is reflected in the release statute 16-104-18 that requires that the patient no longer suffers from an abnormal mental condition ... likely to cause him to be dangerous to himself, to others, or to the community if at large." In another part of the statutes, the older or more archaic term of "mental disease or defect" is used. Legal consultants suggested that legislative intent had to do with the cognitive test of defect of reasoning. Psychologists have suggested that the mental disease or defect phrase is more restrictive to major mental disorders than the broader term of abnormal mental condition. The researcher here used the one most cited by district court judges in trial transcripts and court orders for examination, the abnormal mental condition. Patient Petition for Release The hospital was under no obligation to review for release eligibility unless the patient filed a motion with the court of commitment. The sample of 109 conditional releases represents approximately 10 percent of all hospital release examinations for criminal court review during the time period studied. Most examinations were the result of the patient's legal motion, rather than by team recommendation. In 57

PAGE 65

one conservative year, for example, there were 95 patients reviewed with only 11 having been initiated by the clinical teams. Track II accounted for more self-initiated release hearings, probably because of their better social competency. Psychiatric Recommendations Primary clinical criteria for release eligibility at the Colorado hospital was the expectation that the patient discuss his crime of commitment, understand his legal status and diagnosis, express some remorse for his violent acts, recognize his own danger signals, and plan alternate methods of expression and problem solving. All of these abilities required some minimal level of heal thy, prosocial, interpersonal communication and understanding. Therapeutic relationships with staff and peers were, therefore, crucial to the release process. Prior to administrative and court review, a psychiatrist from within the agency provided a written, formal examination of the patient's mental status and prognosis for future danger. Patients with lifelong mental disabilities requiring the use of antipsychotic medication, were expected to be candid about their symptoms (for example, the frequency and content of hallucinations) and their response to medication (preferences, side effects, improvements). The treatment goal was to teach the patient how to manage his own illness or impulses, with fewer and fewer external controls. The Administrative Review Patient petitions and team recommendations were all reviewed by the hospital's multidisciplinary review board, the Disposition 58

PAGE 66

Committee. Geriatric NGR!s and female NGR!s housed on the general adult civil unit of the hospital also were under the jurisdiction of the Disposition Committee for administrative and clinical review. After the Committee privately reviewed the written history and psychiatric recommendations, the patient was usually interviewed by the group. Rarely were attorneys present, but the patient's primary therapist usually attended to answer questions and support the patient. The board exercised considerable influence within the organization and the courts of commitment, and its legacy dates back to the early 1960s. The board members then included law enforcement officers and politically appointed prominent citizens, whereas today only hospital clinicians sit on the board. As in the beginning, the Committee's official mission is protectionist with a goal of minimizing false negatives, that is, inaccurate predictions of safety. The psychiatric opinion became part of the written Committee packet sent to the superintendent, the hospital's chief executive officer and herself a psychiatrist. The board's evaluation was written by the Committee chairman, a clinical psychologist, who described the patient's psychiatric history, criminal record, course in hospital, and behavior in interview. The Official Agency Recommendation The agency concluded with one official position in the form of a cover letter by the chief executive officer and superintendent, also a psychiatrist. Her recommendation constituted a prediction. The letter to the NGRI court of commitment gave identifying information 59

PAGE 67

and the date of release examination, followed by a standardized opinion: After reviewing the defendant's record and attached reports, it is my opinion that this defendant continues to suffer from a mental disease or defect which is likely to cause him to be dangerous to himself, to others, or to the community in the reasonably foreseeable future (CRS 16-8-115, as amended). Therefore, it is the undersigned's opinion that the defendant is not eligible for release from hospital care and treatment. This example is of an opposition letter, estimated to have been the official agency stance on about 85 percent of all release hearings. When the agency recommended release the courts nearly always concurred. There were only 31 NGRis conditionally released over the hospital's opposition in the six year study period. Adversarial Release Hearings The district attorney in the jurisdiction of the commiting court could contest any release, whether recommended or opposed by the hospital. The D.A.'s practice was to request the court to order a second psychiatric opinion, occasionally a doctor of the prosecutor's choice. The process slowed down and sometimes aborted the release process. When the patient hired an outside psychiatrist the tes-timony was provided directly to the court as independent of the Committee and hospital. Courts generally attended to written reports when no one was recommending release, and heard testimony from expert witnesses when a recommended release was contested. Colorado law defined psychiatric technicians, nurses, teachers, and mental health workers as experts, in addition to psychiatrists and psychologists. It was not uncommon for the officials representing the 60

PAGE 68

hospital position to have members of its own staff on the other side of the courtroom. Jury trials were infrequent, but were more likely when there were conflicting opinions and when the district attorney was opposed to release. The burden of proof was on the patient to convince the court that he was no longer insane or dangerous. Data Sources and Limitations Social and psychiatric histories were abstracted and coded from medical and legal records of the hospital. Preadmission arrest records were obtained by the agency's police department from the Federal Bureau of Investigation (FBI rap sheets) with fingerprints taken on admission. Similar arrest records were obtained for follow-up in September, 1987. Crimes committed during the hospitalization, while in custody or on escape, were usually entered in subsequent NCIC records by the arresting authority, and required close attention to coding dates. Reliability checks were made by examining the dates of patient absences and locations. Presence of in-hospital criminal charges was discovered late in the study, but the variable may well have been a good identifier of those NGRis who were later dangerous. In the small number of cases where no arrest record existed, even though there was a known instant offense and court commitment, the researcher relied on the medical and court records kept by the hospital and mental health centers for aftercare. 61

PAGE 69

Patient Data System Items of criminal and psychiatric history reported in previous research as associated with the assessment and prediction of dangerous behavior were included in the 1980 IFP Patient Data System (PDS). Individual patient records from PDS were used to create a research file for microcomputer use and analyses by SPSS/PC+ software. Selected subjects and variables were transferred to the University,s mainframe for access to SPSS-X software. Names and identifying numbers were eliminated for subsequent aggregate analyses. Hospital Records Outcome data came from follow-up variables kept in individual aftercare records, court reports, arrest records, and from o bser-vations of clinicians, patient families, and other patients. An aftercare coordinating nurse, William Leimeister, was interviewed at several phases of the project for clarification of anecdotal notes and formal reports to the courts. Data were abstracted, coded, and incorporated in the project microcomputer and mainframe files. One major limitation to any study on the prediction of dangerous behavior is the fact that those who are clinically defined as dangerous do not usually get released with opportunity for recidivism. The present sample only captured 31 in six years who were released over hospital opposition. The methodological problem was that there was no similar outcome criterion on which to evaluate the accuracy of predictions on patients who remained under the external controls of the hospital. Those who did not get out of the hospital, with or 62

PAGE 70

without hospital recommendation, limit the generalizations to be made from release samples. Level of Measurement ---Most data in this research were at a categorical level of measurement. There were few continuous measures, and no instru-ments with items of statistically derived weights, available in the literature for direct measurement of danger. The lack of continuous level of measurement not only restricted the prospective identification of high risk offenders, but restricted comparative research on dangerous outcomes. Experts repeatedly report that actuarial and statistical methods of prediction are better than clinical; yet what rudimentary instru-ments there are contain clinical and interpretive items such as "manipulative, affective, or predatory" (Meloy, 1969). The weakest factor in building and contributing to a knowledge base is the problem of operational definitions for comparative study and repli-cation. Validity and Reliability Contrary to researcher expectations, arrest records consistently under reported criminal arrests known to the hospital aftercare nurse. Valid, reliable, and replicable outcome variables to define psychiatric relapse and criminal recidivism, and to distinguish disruptions related to abnormal mental condition, were elusive. Much trial and error, as well as case-by-case judgment, were required before settling on a definition of dangerous behavior after release. Concern with under-63

PAGE 71

reporting was reduced by including all other possible data sources as worthy of verification efforts. Severity of outcome disruptions became a problem for class-ification of behavior as safe or dangerous. Arrest or rehospital-ization for crimes of violence was a straightforward definition of dangerous behavior. Other disruptions resulted in arrest or rehos-pitalization without revocation of conditional release, and some involved potentially violent acts like property destruction or menac-ing. A number of cross tabulations and experimental combinations of disruptive events and legal dispositions (discussed further in Chapter 4) led to the inclusion of all revocations in the definition of danger-ous behavior after release, even though the criminal or psychiatric disruptions were not always assaultive or violent. Arrests for index crimes of violence, whatever the disposition, constituted danger because of the severity of the initial charge of arresting officers. Revocation required psychiatric testimony on the release eligibility criterion, thereby legally classifying the patient as again dangerous. Operational Definitions Outcome variables were designed to describe behavioral events after release in terms that could be replicated on other samples of NGRI patient offenders. Events were further classifed according to the type of external controls used in response to NGRI behaviors. Events of criminal recidivism and psychiatric relapse, and consequent legal dispositions by arrest or rehospitalization, were defined. 64

PAGE 72

Disruption After Release Arrest. Overnight or longer detention by police, including taking the patient into custody for forced psychiatric examination by warrant of arrest from the court. Rehospitalization. Overnight or longer admission to any psychiatric hospital, under civil or criminal statutes. New Crime. Category of criminal offense reflecting charges filed at (re)arrest. Violent= homicide, sexual assault, aggravated assault, armed robbery. Nonviolent= menacing, arson, criminal mischief, burglary, trespass, forgery/fraud, disorderly, traffic, status offenses. Assault. Crimes of violence and assaults on another person, with or without criminal charges filed, involving arrest or rehospitalization. Threats. Menacing with weapons; stated intent to harm, kill, bomb, burn, etc., involving arrest or rehospitalization. Relapse. Psychotic decompensation, disabling symptoms, involving arrest or rehospitalization. Off Meds. Noncompliance with prescribed neuroleptics, generally diagnosed psychotic and chronically mentally ill, involving arrest or rehospitalization. Abuse. Use of alcohol or other nonprescription drug, with resulting disruption in interpersonal, employment, or illegal acts, involving arrest or rehospitalization. Generally in diagnosed substance abusers. 65

PAGE 73

Disposition After Disruption Revocation. Court ordered termination of conditional release, on the ruling that patient is no longer eligible (not dangerous as a result of an abnormal mental condition) to remain on release. The patient is returned to his former NGRI legal status. NGRI. New criminal court commitment on new finding of insanity to criminal charges acquired after release. ITP. New criminal court commitment as incompetent to proceed to criminal charges or proceedings initiated after release. MHC. Mental health commitment under civil 27-10 procedures. Includes voluntary rehospitalizations. Corrections. Any disposition following new criminal charges that involves criminal probation, incarceration, community corrections, deferred prosecution or sentence. Assumes patient accountability and conviction. Prehospital Disruptions Juvenile History. Arrested for criminal behavior before the age of 18, with some evidence of juvenile adjudication, probation, or detention for serious problems with the law Excludes status offenses or acts not considered criminal if commited by an adult (runaway, truancy). Prior Violent Crime. Officially charged and on arrest record, whether convicted with correctional disposition or a mental health diversion. Crimes of violence = homicide,sexual assault, aggravated assault, armed robbery. 66

PAGE 74

Prior Incarceration. Any correctional sentence to jail, prison, or reformatory over 60 days in lock-up. Excludes suspended sen-tences and jail time pending trial on instant offense, and sentences served by probation. Violence of Instant Offense. (Rank order is approximate severity on face validity only, most to least serious.) Homicide Sexual Assault Aggravated Assault Kidnap Armed Robbery Menacing, simple assault Arson Criminal Mischief Burglary Trespass/Theft Forgery/Fraud Nonassaultive Sexual Status Offense Disturbance/Disorderly Resisting Arrest Traffic Arrests Use of Weapon in Instant Offense. In police reports of crime leading to NGRI commitment, use of a knife, gun, other tools for attack or intimidation. Use of Alcohol in the Instant Offense. In police reports or by self-report, the NGRI was under the visible influence of alcohol at the time of the instant offense. Use of Drugs in the Instant Offense. In police reports or by self report, the NGRI was under the influence of nonprescription mood altering chemicals; other than or in addition to, alcohol. Prior Treatment History. History of previous psychiatric inpatient treatment; history of alcohol abuse; history of drug abuse. 67

PAGE 75

In-Hospital Disruptions Stay, Months in hospital from date of admission to date of exit on conditional release. Time. Months in NGRI status only; excluded incompetent time or civil time under definition of "stay" above. Maximum. Number of days accumulated on highest level of security. Medium. Number of days accumulated on medium security. Intermediate. Number of days accumulated on (locked) wards. Minimum. Number of days accumulated on IFP (open) wards. Special Leave. Number of absences to court or corrections. Days Authorized Leave. Number of days accumulated on special leave to custody of court or correctional agencys. Passes. Number of times placed on overnight or longer pass status. Days on Pass. Number of accumulated days on pass. BMT. Number of Behavioral Management Transfers to higher security level. N Escape. Number of times absent without authority. Escape Days. Number of accumulated days on status. Treatment Track. Wards 5, 7, 9, 11, 79 = Track I. Wards 2, 10, 12, GW 5,11 = Track II. Temporary locations receded. Diagnostic Categories. Psychotic diagnoses = schizophrenia, organic psychoses, bipolar disorders, atypical psychoses, paranoid states. Substance diagnoses = substance induced organic, alcohol abuse or dependence, drug abuse or dependence. Other nonpsychotic 68

PAGE 76

= personality disorders, anxiety neuroses, organic personality syndrome, mental retardation, adjustment reactions, sexual deviation. Coded first in DSM III nomenclature to five digits, then categorized. Interval Level (above) Variables. Subsequent collapse of time intervals for absences, length of stay, time at risk, into logical intervals. Median used for some dichotomies. Legal Dangerousness Scale. New York Department of Mental Hygiene, no copyright, 1974 Cocozza & Steadman; Juvenile record = 8 points; Previous incarceration = 4 points; Prior violent crime = 2 points; Crimes 1 to 11 serious instant offense = 1 point. Statistics The hypothesis that clinical assessment and prediction would be more accurate than chance distribution of risk was first tested by nonparametric methods of dichotomous predictions and dangerous outcomes. Independent variables (predictor items from patient histories) were taken from findings reported to be associated with dangerous behavior. Interpretations of types of errors were made according to the policy perspectives of the civil libertarian and public protectionist models. Distributions were analyzed for total accuracy, as well as by Relative Improvement Over Chance (RIOC) to compare accuracy and type or direction of errors (i.e. inaccurate predictions of safety or inaccurate classification as dangerous). Selection factors were analyzed for optimal cutting scores for classification of high and low risk. The selection percent in RIOC computations is the proportion 69

PAGE 77

of the sample captured by the item or score under consideration. The formula for RIOC is dependent upon the known frequencies in the distribution. RELATIVE IMPROVEMENT OVER CHANCE = ACHIEVED ACCURATE RANDOM MAX POSSIBLE RANDOM The maximum possible total correct predictions, in the two-by-two contingency tables used here, is constrained by the actual row and column totals. Total achieved accurate predictions are the true positives plus true negatives. Percent total accuracy is the sum of accurate predictions divided by N (109) times 100. Maximum possible accurate predictions are constrained by actual outcomes, in this case row totals. That is, the calculated theoretical "max possible" can only include the minimum number of patients who were actually dangerous after release. Bayesian probabilities are a form of statistical inference based on a theorum of equality among prior events, proposed by an English clergyman in 1763, Thomas Bayes (Iversen, 1964). In the SPSS-X discriminant analysis, the prior probabilities were set equal to the proportional distribution, that is, .60 = SAFE and .20 = DANGEROUS. The resulting functions were then compared to an equal probability model with a .50 SAFE and .50 DANGEROUS proportional distribution. The research task was to find the variables which (in retros-pect) best discriminated the dangerous from the safe patients at follow-up. Because outcome variables were highly skewed, as in low base rate behaviors, initial analyses involved a number of statistical 70

PAGE 78

approaches. Discriminant function analysis (Klassen & O'Connor, 1988; Klecka, 1984; Steadman & Morrissey, 1981) was selected as the most appropriate statistic for the data set. A similar analysis for distinguishing the groups opposed for release by a prediction of danger was applied. The stepwise method for deriving the equation, which interactively discriminates the classifications of dangerous patients, eliminates extraneous variables. An F-ratio of 1.00 was selected to be broadly inclusive and to identify variables with minimal explanation for differences between SAFE and DANGEROUS groups. A secondary goal was to determine if the patients predicted by clinical and agency officials to be dangerous, that is, the group opposed for release, could also be distinguished by other i terns of criminal or psychiatric disruption in history. Items were entered in logical combination as 1) sociodemographic variables of _education, age, ethnicity, employment, and marital history; 2) clinical variables of diagnosis, program, alcohol and drug abuse, security, passes, and the like; and 3) criminal variables of juvenile record, prior violent crime, incarceration, use of a weapon, and probation history. Discriminant analyses were first applied separately to the three categories of items to determine the optimal discrimination of outcome groups and prediction groups. The wide range of time with opportunity to recidivate or relapse was controlled for by entering each subject's release date subtracted from the date of followup. The six year range for releases was expected to contribute significant differences in rates of disruption. 71

PAGE 79

Variable reduction in the final discriminant function equation selected the most efficient number and type of variables for distin-guishing the dangerous from the safe. Confidence levels vary, according to the specific technique, but in general have an upper limit of .05 for consideration in predictive equations. Relative Improvement Over Chance (RIOC) is a somewhat absolute value in relationship to the distribution being analyzed for predictive accuracy, so the critical region is positive and directional. RIOC findings with this sample ranged from a negative (the information was worse than chance) to an 80 % improvement over chance. The end product of the various statistical approaches were two sets of tables, one pair for the dependent variable of DANGER and one pair for the dependent variable of OPPOSE. The pairs of tables are for comparison of the Bayesian (actual) probability and equal probability models. The classification results obtained from the discriminant function for each of the four tables yielded more precise information about predictor variables than RIOC. Differences between and among the technical models were evaluated for utility and value in comparison and replication. 72

PAGE 80

CHAPTER 4 RESULTS Routine nonparametric tests for association among pre-hospital, in-hospital, and post-hospital disruptions provided the direction for subsequent application of Relative Improvement Over Chance (RiOC) and discriminant function analyses with equal and Bayesian probabilities. An operational definition for dangerous behavior after release was deduced from the following initial analyses of outcome data. The proposition that agency opposition to release, that is, a classification as dangerous, would distinguish the subsequently dangerous was confirmed by three different statistical approaches. Definition of Dangerous Disruption Over half (51.4 %) of the sample had no disruptions leading to arrest or rehospitalization. The social control mechanisms of arrest and rehospitalization were closely related; that is, if the patient avoided arrest, he was also likely to avoid rehospitalization. Of 69 not arrested, 56 were also not rehospitalized (X2 = 23.48, df 1, p. < 0001). The events leading to social control, and the legal dispositions after disruptive events, were analyzed for operational definitions of dangerous and nondangerous disruptions. In Table 4.1 initial frequencies of unclassified events provided the foundation for organizing the data according to study propositions.

PAGE 81

Table 4.1. Involving Arrest or Rehospitalization after Conditional Release (n = 109). Arrested Rehospitalized 40 (36.7%) 39 (35.8%) Neither 56 (51.4 f.) Both 26 (23.9 %) Disruptions Assaultive Threats to ham Off medication Psychotic relapse Drug abuse 23 (21.1 %) 21 (19.3 f.) 15 (13.8 %) 20 (18.3 f.) 23 (21.1 f.) Note: S may appear in more than one disruption. Crime Category Crime of violence 15 (13.8 f.) Potential violence 12 (11.0.%) Nonviolent crime 13 (11.9 f.) No new crime 69 (63.3 %) Note: S appears once in most serious crime category. No homicides or attempts were included in the criminal charges after release, but there were five sexual assaults, seven aggravated assaults, and three armed robberies. Potential violence included four menacing or simple assaults, one arson, four burglaries, and three property destruction (mischief) charges. Remaining criminal charges were for trespass, fraudulent use of credit cards, disturbance and the like. Two were arrested for violations related to eluding police and reckless driving. Criminal conduct also included arrests for violation of court ordered conditions of release, and for minor disturbances associated with psychotic decompensation. Except when revocation occurred by 74

PAGE 82

declaring the patient again legally dangerous (n=12), only patients alleged to have committed crimes of violence (sexual assault, agg-ravated assault, or armed robbery) were classified as dangerous. The one new NGRI (predicted safe) was also revoked and in the classification of dangerous. All of the assaults precipitating arrest or hospital confinement were captured with the definition of dangerous disruption. Various crosstabulations and nonparametric tests for association were applied to the events of disruption and to the dispositions taken by agents of social control. As with highly correlated arrest and rehospitalization, mental health dispositions were found to occur in response to psychotic behaviors. More than one type of disposition may also have been used in response to the same event. The univariate frequencies in Table 4.2 include such duplications. Table 4.2. Disposition of Disruptive Events Involving Arrest or Rehospitalization after Conditional Release (n = 109). Revocation of C.R. New NGRI Incompetent for trial Civil or Voluntary Correctional 12 (11.0 'l.) 1 ( 0.9 'l.) 3 ( 2.8 'l.) 27 (24.8 'l.) 27 (24.8 'l.) Note: Disruptive Ss (53) account for more than one disposition. Nearly half of the sample had significant disruption, but not all disruptive events involved assaultive behavior, nor did serious crimes of violence always lead to revocation of conditional release. Six types of disruptive behavior and five types of disposition were 75

PAGE 83

analyzed with bivariate distributions. Of 15 crimes of violence, 10 (67 %) were accounted for by Track II patients. Of 12 revocations of condi tiona! release, 8 (67 'l.) were accounted for by Track I patients. Nonpsychotic Track II patients thus were more likely to have dangerous behavior handled by correctional disposition. Combining these various events and dispositions led to the following two-by-three contingency table used for initial descriptive statistics. Each patient only appears once in the classification of outcomes, according to his agency recommendation (or opposition) for release. Column 3, Dangerous Disruption, includes all patients with crimes of violence and/or revocation of conditional release. Table 4.3. Types of Disruptions Involving Arrest or Rehospitalization after Conditional Release by Hospital Prediction (n=109). No Nondangerous Dangerous Disruption Disruption Disruption Predicted Safe 43 (55.1 %) 24 (30.8 %) 11 (14.1 '1) Predicted Danger 13 ( 41.9 'f.) 7 (22.6 7.) 11 (35.5 7.) 56 (51. 4 7.) 31 (28.4 %) 22 (20.2 1.) X2 = 6.30, df 2, p = .0429 78 31 109 In Table 4.3 and for all further statistical analyses, patients were counted only once in the outcome classification of NO DISRUPTION, NONDANGEROUS DISRUPTION, or DANGEROUS DISRUPTION. 76

PAGE 84

In attempts to use the most information possible, but in ways that best enhanced predictive accuracy in low base rate behavior, the three outcome groups were then dichotomized by combining the NO DANGER and NONDANGEROUS groups as SAFE (n = 67), for RIOC and discriminant function analysis. This yielded about an 60/20 split in the dependent variable, driving the remaining calculations. The 22 patients with DANGEROUS DISRUPTION included 6 (16 % of 44) revocationS from Track I and 4 (6.1 'f, of 65) from Track II. Track I had 61 % with no new crimes and Track II had 65 %, not a significant difference. Crimes of violence were twice as likely for Track II patients. The chronicity of mental illness was reflected in higher rates of hospitalization and revocation for Track I, and more use of correctional dispositions for Track II. Statistical Analyses Improvement over chance is a function of the maximum possible correct classifications and the expected frequencies (random or chance distribution). In any particular study the value or percent RIOC depends on the size of the sample, the accuracy rate or achieved correct, and the relative improvement in accuracy that the item or score adds. The marginal totals that constrained the initial analysis were 67 SAFE AND 22 DANGEROUS, or an 80/20 proportion-al split. The RIOC statistic is a contingency table analysis of categorical distributions, so technically the value is not dependent on continuous measures. Individual variables were each tested by RIOC in two-by-77

PAGE 85

two tables. RIOC was also used to determine the optimal cutting score for yielding the fewest erroneous classifications (predictions) with continuous variables. Reduction of the number of variables to the most parsimonious explanations, and the elimination of redundant measures, was an objective of all statistical manipulations. Dangerous Outcomes RIOC Results with the dependent variable DANGEROUS DISRUPTION are shown in Table 4.4 where most variables were included in the categories of 1) sociodemographic 2) criminal history and 3) clinical course items. Classifications were similar to the Klassen and O'Connor (1988) discriminant analysis with 67 potential predictors. The organization of information in Table 4.4 used a format from Loeber and Dishion (1983) to emphasize the errors (false negatives and false positives) that the model tries to minimize. Percent of the base rate (% BR) is the actual rate of dangerous behavior; percent of the selection ratio (% SR) represents the proportion of the total sample who would have to be detained to prevent dangerous behavior, that is, classified as high risk on the independent (predictor) variable. Individual items at a continuous level of measurement were analyzed by RIOC to determine the best cutting scores for high and low risk group assignments reported in Tables 4.4 and 4.5. More than 2 COURT leaves, less than 2 PASSES, any BMT (behavioral management transfer), and any hospital ESCAPE were found by RIOC to improve predictive power significantly. These in-hospital disruptions were used in unweighted combination as a composite variable (REBEL) in subsequent discriminant analysis. 78

PAGE 86

Table 4.4 Accuracy of Predictions of Dangerous Behavior After Conditional Release (n=109). False False True Positives True Negatives % % Pos. n % Neg. n % Corr. o/oBR %SR RIOC Sociodemographic Age < 30 yrs. 10 26 23.8 61 12 11.0 65.1 20.2 33.0 36.4 Minority 12 26 23.8 61 10 9.1 67.0 20.2 34.9 28.6 Educ. < h.s. 13 37 33.9 50 9 8.3 57.8 20.2 45.9 25.0 Never Married 10 39 35.8 48 12 11.0 53.2 20.2 45.0 0.0 Longest Job < 3 yrs. 20 61 55.9 26 2 1.8 42.2 20.2 74.3 66.7 Criminal History Juvenile Record 14 34 31.1 53 8 7.3 61.5 20.2 44.0 33.3 -.;a Prior Violence 10 22 20.2 65 12 11.0 68.8 20.2 29.4 25.0 ..0 Incarceration 13 28 25.7 59 9 8.3 66.1 20.2 37.6 35.7 Violent Instant Off. 13 53 48.6 34 9 8.3 56.9 20.2 60.6 35.7 Use Alcohol/Drugs 14 33 30.3 54 8 7.3 62.4 20.2 43.1 38.5 LDS > 5 15 40 36.7 47 7 6.4 56.9 20.2 50.5 36.4 Weapon Instant Off. 10 55 50.4 32 12 11.0 38.5 20.2 59.6 -33.3 Clinical Course Nonpsychotic Dx 15 50 45.9 37 7 6.4 47.7 20.2 59.6 22.2 High Security 11 18 16.5 69 11 10.1 73.4 20.2 26.6 31.3 Escapes 11 33 30.3 54 11 10.1 59.6 20.2 40.4 15.4 > 25% Lockup 21 63 57.8 24 1 0.9 41.3 20.2 77.1 80.0 > 2 Court Trips 15 41 37.6 46 7 6.4 56.0 20.2 51.4 36.4 < 2 Passes 17 45 41.3 42 5 4.5 54.1 20.2 56.9 44.4 BM Transfers 16 46 42.2 41 6 5.5 20.2 56.9 33.3 Rebel Score > 2 18 48 44.0 39 4 3.7 52.3 20.2 60.6 55.6 Note: BR = % Baserate; SR = % Selection Ratio; RIOC = Relative Improvement Over Chance.

PAGE 87

Table 4.5. Hospital Recommendations For or Against Release by In-Hospital Disruptions as Predictors (n=109). False False True Positives True Negatives % % Disru:Qtions Pos. n % Neg. n % Corr. %BR 'YaSR RIOC More than 2 Court Trips 22 34 31.2 44 9 8.3 60.5 28.4 51.4 40.0 More than 25% Lockup 28 56 51.4 22 3 2.8 45.9 28.4 77.0 57.1 BM Transfers 23 39 35.8 39 8 7.3 56.9 28.4 56.9 38.5 CD Less than 2 Passes 24 31 28.4 47 7 6.4 65.1 28.4 50.5 53.3 0 Program II 24 41 37.6 37 7 6.4 55.9 28.4 59.6 46.2 Nonpsychotic 18 26 23.8 52 13 11.9 64.2 28.4 40.3 27.8 Escapes 19 25 22.9 53 12 11.0 66.1 28.4 40.4 33.3 High Security 19 10 9.2 68 12 11.0 79.8 28.4 26.6 52.4 Note: BR = Baserate; SR = % Selection Ratio; RIOC = Relative Improvement Over Chance.

PAGE 88

Sociodemographic variables, when measured by RIOC, were directional, but not at particularly high accuracy rates. Minority groups had more trouble after release, but were not more likely to be opposed for release. A stable job history was a low risk indicator, with two false negatives, but the selection rate classified nearly three-fourths (% SR) as high risk for DANGER. Marital status was not a helpful discriminator when used alone against dangerous outcomes. Knowledge of a patient's prior violent crime improved the accuracy with an RIOC of 25% and a total accuracy rate of 68.8% The number of false positives was reduced to 22, but false negatives were 12 (11 % is the highest rate of false negatives reported). Prior violence and other criminal items were subsequently tested in a composite variable by discriminant function analysis. Clinical variables helped differentiate the dangerous by accuracy rates and RIOC, particularly items related to in-hospital security classifications. Patients released from maximum or medium security had a greater probability of DANGEROUS outcomes than patients released from intermediate or minimum security. The proportion of the length of stay spent on maximum and medium security wards was calculated in a variable named LOCKUP. It was first treated as a continuous variable for RIOC analysis of a cutting score. LOCKUP greater than 25% captured all but one dangerous outcome and yielded an 80% RIOC, but at the expense of 63 false positives. Patient disruption during hospitalization was operationalized by the REBEL composite variable, which yielded only 81

PAGE 89

four false negatives and produced an RIOC of 55.6 percent better than chance on the DANGER outcome variable. Disruption in hospital was also found in the next analysis to be significantly related to hospital opposition. Agency Opposition RIOC Patients opposed (31) and patients recommended for release (78) constituted the dependent variable for testing the second research proposition with results in Table 4.5. No significant differences were found for OPPOSITION on the variables of ethnic minority, education, age, marital, job or criminal history. In-hospital disruptions started to appear to be important to the clinical prediction variable when using RIOC to analyze patients' conformity to the graduated treatment program. The idea of "going through the system" came to be operationalized by the absense of regressive transfers, hospital escapes, high security classification at the time of release, and failure to have been tested with passes in the community before conditional release. Individual items were related, and summing them in a variable called SYSTEM, led to enough improvement over chance to include the items in subsequent analyses. To code for directional purposes in tabulating, the SYSTEM variable was transformed to one called REBEL, believed to describe nonconformity to agency expectations. REBEL distributions allowed False negatives to be minimized to one with low REBEL scores, at the expense of 62 false positives. Raising the score to over 2 optimized errors to 5 false negatives and only 40 false positives. The 82

PAGE 90

fewest number of erroneous predictions of safety (recommendation for release) in this scheme (3) came from percent hospital career in LOCKUP, but again resulted in a 77'1. selection ratio and 56 false positives. LOCKUP was the proportion of STAY spent by the patient on high security level wards. All values of RIOC are relative and dependent on the 28% base rate of OPPOSITION in the actual agency classification of this sample. Discriminant Function Analysis of Dangerous Outcomes To evaluate the proposition that civil libertarian models would work to minimize false positives, a Bayesian model of actual probabilities was derived from the sets of sociodemographic variables, criminal history items, and clinical variables. The stepwise discriminant procedure eliminated a number of variables from the equation. Removed were AGE, STAY, RISK, LDS, DIAGNOSIS, WEAPON, INSTANT OFFENSE, ALCOHOL, DRUG, EDUCATION, MARITAL, and REBEL. The first pair of comparisons of the two probability models, included the same six variables in the final two equations and discriminant functions. In Table 4.6 the discriminant function scores assigned group membership(s) in the "Predicted" columns. Significant variables and their standardized canonical coefficients for distinguishing DANGER were LOCKUP .87; MINORITY .34; PROGRAM -.47; JOB .30; PRIOR .24; and OPPOSED .22. In the first probability model (50:50), 35 are classified as high risk for danger, and about half of those are correctly captured with the predictor i terns in the discriminant function. As well, the 83

PAGE 91

scheme captures 81.8% of the 22 actual dangerous outcomes. In the Bayesian model, only 12 get classified as dangerous, but 67% of those were dangerous. Table 4.6. Predicted and Actual Group Membership for Dangerous Outcomes with Equal and Bayesian Probability Models. Equal Probability Predicted Safe Danger Total Outcome Safe 70 17 87 Danger 4 18 22 Total 74 35 109 % Correct = 80.7% (88/109) Total Errors = 21 False Pos. = 80.9% (17/21) False Neg. = 19.1'1 (4/24) RIOC = 73.3% Bayesian Probability Predicted Safe Danger Total Outcome Safe 83 4 87 Danger 14 8 22 Total 97 12 109 % Correct = 83.5'1 (91/109) Total Errors = 18 False Pas. = 22.2'1 (4/18) False Neg. = 77.8% (14/18) RIOC = 60.0% The proportion of the patient's hospital stay spent on high levels of maximum and medium security (LOCKUP) had the strongest explanatory power for the variance in outcomes. In combination with the other five variables that met the threshold criterion for inclusion, hospital OPPOSITION shared but did not dominate the discriminant function. Analyses of errors and the type of errors from the libertarian and the protectionist perspectives were contrasted in Table 4.6. Minimization of false positives (4) was best achieved by a Bayesian model. In contrast, minimization of false negatives (4) was best 84

PAGE 92

achieved by an equal probability model favored by protectionists. True to previous accounts, there is a systematic tradeoff in the direction of errors when minimizing one type over the other. Using actual probabilities of 80:20 clearly reduced the libertarian concerns for unnecessary confinement on the basis of a danger classification. Discriminant Function Analyses of Opposition Predictor variables and their canonical coefficients for inclusion in the equation for OPPOSE were REBEL .99; PROGRAM .38; ABUSER -38; and JOB .-26, for the equal probability equation. The discriminators yielded a 78.9 % total accuracy rate. The Bayesian prior probability model removed JOB from the equation and achieved an 80.7 %total accuracy rate. As in the DANGER analyses, it is shown in Table 4.7 that when the false positives were minimized to 7, the Bayesian model produced 14 false negatives. Likewise, the equal probability model would erroneously classify 16 patients as dangerous in order to reduce false negatives to 7. Any psychiatric diagnosis of alcohol or drug abuse or depen dence was selected as an ABUSER variable and found to distinguish between patients OPPOSED for release and patients recommended for release. ABUSER, however, was removed from the equation for predicting DANGER. 85

PAGE 93

Table 4.7. Predicted and Actual Group Membership for Hospital Opposition with Equal and Bayesian Probability Models. Equal Probability Bayesian Probability Predicted Predicted No Yes Total No Yes Total Oppose Oppose No 62 16 78 No 71 7 78 Yes 7 24 31 Yes 14 17 31 Total 69 40 109 Total 85 24 109 % Correct = 78.9% (86/109) % Correct = 80.7% (88/109) Total Errors = 23 Total Errors = 21 False Pos. = 69.6% (16/23) False Pos. = 33.3% (7/21) False Neg. = 30.4% (7/23) False Neg. = 66.7% (14/21) RIOC = 65.0% RIOC = 58.8% Summary of Findings The PROGRAM variable was included in the final discriminant function analysis for both DANGER and HOSPITAL OPPOSITION groups, each group defined as "dangerous" in this study. The PROGRAM distinction was found to be somewhat of a composite variable in itself. Significant differences were found between Tracks I and II in diagnosis, marital, educational, criminal record, alcohol and drug abuse, and in-hospital disruptions, described under the Sample. The shared variance of those variables may wen have been captured by PROGRAM, explaining why a number of variables thought to be important were removed by earlier discriminant analyses. The LOCKUP variable was also correlated with PROGRAM. Track II patients spent proportionately more time on high security wards. The underlying construct being measured may have been 86

PAGE 94

personality, assessed danger in hospital, "criminality" rather than abnormal mental condition, nonresponsiveness to treatment, or none of the above. Whatever clinical or management criteria contributed to program assignment probably contributed to the operational definitions of dangerous behavior after release, as well as the classification of danger by agency opposition to release. Subsequent decisions for revocation (12) or insanity (1) were often made by the same agency's clinical administrators and psychiatrists. The REBEL composite measure of in-hospital disruption (court trips, behavioral management transfers, escapes, and no passes) was strongly associated with agency OPPOSITION. Surprisingly, however, in-hospital disruption was not predictive of subsequent DANGEROUS DISRUPTION. The REBEL variable was removed from the equations for DANGER under both prior probability conditions and did not discriminate the dangerous from the safe groups. And finally, OPPOSITION did enter the prediction of DANGER classification, but not weighted as powerfully as the other five discriminators. The effects of experimentally omitting one variable and then another from the discriminant function confirmed the value of the technique of combining variables to explain the most variance. When the OPPOSE prediction included PROGRAM, REBEL, AND ABUSER, the total correct classification rate was 78.9 % under equal probability conditions. Removing PROGRAM reduced accuracy to 71.6 %. Including PROGRAM and removing REBEL dropped accuracy to 55.9 % and the discriminant function also removed ABUSER. It would appear that if decision makers had only one i tern of information to make 87

PAGE 95

recommendations to the court, disruption in hospital (REBEL) would be the most helpful. If evaluators of those predictions had only one i tern of information to identify the DANGEROUS after release, security classification in hospital (LOCKUP) would be most helpful. 88

PAGE 96

CHAPTER 5 DISCUSSION The simplistic notion of first generation research was to predict everyone safe in light of the high rate of erroneous predictions of danger. The idea has some statistical substance as well. If all 109 patients had been predicted safe, the error rate would have been the 22 (20%) who were subsequently dangerous. In actual fact, all 109 NGRis in this sample were indeed predicted safe by someone, including the judges and juries who released 31 patients over the objections of the agency. As in the Colorado (1977) study of 111 NGRis also predicted to be safe, the present sample had significant trouble after release in nearly half of the cases. Arrest and rehospitalization were defined as external controls exercised in response to social disruption. The base rate for this sample is 48.6 percent when both nondangerous and dangerous disruptions are included. The further distinction made here to define only 22 as dangerous relied on the legal definition for revocation, and on FBI Index crimes of violence. Public protectionists have scientific evidence from this study, and Steadman and Morrissey (1981), that an equal probability model is one way to best minimize false negatives. No combination of variables attempted here managed to reduce false negatives to zero, outside of predicting every patient to be dangerous (a false positive ratio of 87/22 and total accuracy of only 20%). The best cases for a

PAGE 97

Bayesian model consistently reduce false positives at the expense of false negatives of 10, 12, or more. Caution is suggested before cross validation studies can attempt replication of specific predictor variables. As pointed out by Klassen and O'Connor (1988) these statistical techniques produce "tailor-made" equations for each particular set of data. There is thus an inherent instability in low base rate behavior with relatively small samples. On the other hand, it may be helpful to have program-specific and sample-specific indicators for risk assignment of future NGRI releases from this and similar agencies. Uncertainty about what PROGRAM assignment is actually measuring is acknowledged, but whatever it may be, treatment assignment is predictive here of both hospital OPPOSITION to release and DANGEROUS DISRUPTION after release. In-hospital disruption as measured by behavioral management transfers, escapes, high security at release, and no passes (REBEL) assured that the agency would oppose release on the danger criterion. Perhaps level of social functioning (premorbid life style) and level of functioning in the hospital are related to treatability, rather than to dangerous characteristics. Some may simply start out in a low risk group with less need for treatment than the highly dangerous. They may proceed without disruption through the graduated security program, using passes for community re-entry, and meeting SYSTEM expectations. They may not be REBELS, but they may not have been dangerous either. As well, institutions are notorious for exaggerating the very behavior they are designed to control (Sykes, 90

PAGE 98

Goffman, Toch). If some patients start out as REBELS, or their experience in program contributes to becoming REBELS, these findings suggest they are unlikely to be released with the hospital's blessing. The REBEL variable did not remain in the discriminant function analyses for dangerous outcomes, being replaced with a different combination of predictor variables (ABUSER, JOB). One explanation is that OPPOSE itself may be a function of the items that were used in composite variables. Thus when OPPOSE entered the prediction formula, individual distributions on the variables became rearranged for the optimal discrimination of the safe and the dangerous. The strength of LOCKUP for DANGEROUS DISRUPTION may also have some colinear relationship with items in REBEL. Base Rates and Probability Models This base rate approaches the equal probability model that Gordon (1982) suggests is reasonable for high risk, homogeneous groups of known offenders. He contends that the Koppin and Kozel samples, and his Patuxent group, were much higher risk than ordinary citizens. His (1977) solution to high rates of false positives was to add 111 nuns (expected safe outcomes) to the Colorado sample, a humorous lesson in heterogenity. Gordon says that whenever material becomes more homogeneous, corrrelations are lowered. He asks "Why should dangerousness be an exception?" (1982, p. 298). Gordon continues the analysis by pointing out the seeming paJ;"adox that as predictive values approach the mid-range probability 91

PAGE 99

of .50, outcomes actually become more heterogeneous. The proportional probabilities set for discriminant analysis, and the marginal (actual) distributions used in RIOC, were attempts to account for skewed outcomes. And in Gordon's review of the few empirical studies, and in this present research, outcome criteria for danger are dichotomous or zero-one probabilities. The classification is acceptable after the fact; but when predicted, danger is not an all-or-none probability of 0 or 100 percent. The selection ratios calculated with various high risk indicators constitute a statement of sample probability, not the same as saying that an individual member of the high risk group has a precise probability of being dangerous. In this sample of 109, hardly anyone had a zero probability, as with the offender populations Gordon studied in his policy analysis. On the other band, when making decisions to release or to detain, the decision does become dichotomous. The offender either is conditionally released (or paroled, or allowed bail, etc.) or he is confined as dangerous. Likewise, until scales or schemes for distinguishing severity of dangerous outcomes are developed, he is either dangerous or safe at follow-up. Explicit and Replicable Criteria One of the Morse (1983) objectives for further scientific study called for more precise and validated criteria for classifying danger. Discriminant function analyses probably is the best technique used so far in the prediction literature, to deal with both kinds of error, and the danger-safe dichotomy. A broad definition of danger was used in the revocation of conditional release in a few cases here. 92

PAGE 100

Disruptive events resulting in other methods of rehospitalization, classified here as nondangerous, did include criminal conduct that could be defined as dangerous in a broader definition. Psychiatric relapse in a chronic schizophrenic, known to be assaultive when off medication, was declared to be imminently dangerous in 27 cases of civil rehospitalization. Some of those cases were classified here as NONDANGEROUS DISRUPTION. The defini-tion for the 22 DANGEROUS is straightforward and replicable by other states with similar statutes (California, Illinois, Oregon). The other kinds of disruption are probably less well-defined and include both criminal recidivism and psychiatric relapse. Using arrest and or rehospitalization for 24 hours or more does help operationalize. Increased Degree and Probability of Harm The second Morse objective called for research to substantiate preventive confinement on the basis of danger predictions. The concern reflects the libertarian concern for false positives. Several cases from that cell were researched and found to have had nondangerous disruption, suggesting that they were not totally misclassified. The best ratios found here of false positives to true positives were in the range of 1.6 to 1.0. Bayesian models yielded one contingency with 85 '.t true positives (accurate identification of danger) but at the expense of 16 false negatives. The Floud and Young (1982) analysis took a different perspective, and discussed the validity of retributive justice and the fact that samples are known to have been responsible for dangerous behavior before being considered for high risk classification. Their 93

PAGE 101

concept of "willful harm" implies a culpability and deliberate choice on the part of the offender. Only one subsequent criminal act among this sample led to a finding of NGRI, suggesting that most former patients were being held accountable in Colorado courts. The false negative problem is believed here to be less conducive to public policy debate than the false positive problem. The latter can be approached as an economic cost of unnecessary hospitalization, whereas placing a dollar benefit on a dangerous act (false negative) is politically unpopular. The lay public would say that one erroneous prediction of safety would be unacceptable, even if given the information of an approximate $500,000 cost for an average NGRI stay ($227 per day for 5 years). The conservative, protectionist perspective and the equal pro ba bili ty model here did the best job of minimizing false negatives. Research Relevant Evidence The third Morse objective was to admit only that evidence that research demonstrates is relevant. Recent reports that attempted to assess situational variables included many interpersonal and family conditions that would be difficult to use as rationale for preventive confinement (relationships with mother, sexual practices, combat experience). Equally problematic in the psychiatric populations would be continuing confinement because of age, sex, race, economic status, or criminal record. Many of these predictor variables are fixed, and no program of treatment has been specific to mitigating such histories. 94

PAGE 102

The psychopathic diagnosis was not studied in the present sample because so few were formally diagnosed. Legislators periodically include and exclude personality disorders from the relevant statutes, which is probably as close to consensus as psychiatrists ever get. Changes in insanity laws in 1983 are believed to have been restrictive and to have discouraged a finding of insanity for personality disorders. Manic-depressives had the shortest lengths of stay (80 'l. of stays under one year) and length of stay was not related to danger predictions or dangerous outcomes. Not enough empirical evidence has been reported for diagnosis to be predictive of danger, but more recent studies have asserted a lot of psychodynamic and diagnostic premises (Rosenberg, et al., 1988). Reformers from both sides of the argument in the early 1980s agreed that psychiatric testimony often became tangential, over inclusive, and irrelevant. Pogrebin and Poole (1986) found institutional disruption of prison inmates to be strongly related to parole board decisions. The present study also found significant relationships between in-hospital disruption and hospital opposition and subsequent dangerous outcomes. Institutions now provide some semblance of due process and advocacy for quasi-legal disciplinary procedures, so it would seem that behavior in the institutions would qualify as evidence demonstrated by research to be related to the criterion validity. Programs to Reduce Propensity for Violence Conformity to the treatment model of graduated reduction in external controls did indicate a favorable recommendation for release 95

PAGE 103

and a better pro ba bili ty of safe behavior after release. Going through the system was uninterrupted for only 16 % of this sample. The number of disruptive events among the NGRis who continue to be confined is unknown, but this sample had its share of behavioral transfers, escapes, court leaves, and high security. PASS days were not significantly related to danger when tested separately, but in the unweighted composite variables of SYSTEM and REBEL, improved discrimination on OPPOSITION and DANGEROUS DISRUPTION was achieved. Insight-oriented psychotherapy permeates both programs and the philosophy of much of the professional staff in the study setting. The implied release criterion that the patient be able to candidly discuss the instant offense, express some remorse, recognize his mental illness, and plan other modes of problem solving, reflects that philosophy. It is not known whether those verbal skills do, in fact, reduce the individual's propensity for dangerous behavior. Self-help programs of Recovery, A.A. and N.A. added another dimension to the treatment of many patients in this sample, making replication more, rather than less, possible in other settings. Track II programs were modeled after therapeutic community systems with character disorders and addicts. This study did not evaluate the degree or type of "dangerousness" at the beginning of treatment, so treatment effects were not studied. Provision of Thorough Review of Confinement Morse (1983) may have had the first generation research in mind when he called for thorough review of continued confinement. The 96

PAGE 104

older studies were not limited to specific samples and individually assessed as dangerous. The present study may be the first one of formal clinical predictions of danger on a legal criterion with a sample of criminally insane. The sample was small, but came from one of the larger forensic units in the country. Data retrei val covered six years of releases in order to obtain a sample larger than 100. Release rates just kept pace with a low commitment rate during the study period. If Morse was thinking of automatic, statutory review on a regular basis, then a utilization review model might be helpful. Asking whether the patient has achieved maximum benefit might open planners and policy makers to alternative, less restrictive placements. If legislative changes were to be seriously considered, and it nearly always takes a crisis in public confidence for change to be considered, the danger criterion itself should be studied by policy analysts. The Colorado statute implies a certain causality, as in the older product tests, that makes dangerous behavior a function of mental disease or defect (abnormal mental condition). The definition of dangerous disruption after release ended up capturing at least two very different types of danger in the 22 outcomes, relevant to mental health policy. Track I revocations tended to be decompensated, acutely psychotic, and if disruptive in the community, more likely to be nonviolent. In contrast, Track II patients committed serious assaultive crimes, and were much less likely to be revoked. 97

PAGE 105

It was as if the definition of outcome DANGER, unplanned in the design, washed out what may have been distinct program differences. Further study might take a clinical case approach to the nondangerous disruptions, handled by civil short-term commitment, to determine if similarly dangerous outcomes were inappropriately classified. If less time could be devoted to the paper compliance of treatment plan reviews as they are now implemented, serious review of the actual necessity for indeterminate inpatient treatment might take place. Annual reports to the Court on the abnormal mental condition, and how it makes the patient likely to be dangerous in the foreseeable future, might meet the last objective of Morse (1983). 98

PAGE 106

BIBLIOGRAPHY Addington v. Texas, 441 SCt 418 (1979). American Psychiatric Association. (1983). Diagnostic and statistical manual of mental disorders (3rd ed. rev.). Washington, DC Barefoot v. Estelle, 103 SCt 3383 (1983). Baxstrom v. Herold, 383 SCt 107 (1966). Braley, N. B. (1981). Estelle v. Smith and psychiatric testimony: New limits on predicting future dangerousness. Baylor Law Review, 33, 1015-1034. Bromley, J. C. (1980, January). 36-40. The RTD slayer. Denver Monthly, Bromley, J. C. (1980, June). Patients' rights vs public safety. Denver Monthly, 67-68. Callahan, L., Mayer, C., & Steadman, H. J. (1987, Jan.-Feb.). Insanity defense reform in the United States: Post-Hinckley. MentalPhysical Law Review, 54-59. Cleckley, H. (1964). The mask of sanity (4th ed.). Saint Louis: Mosby. Cocozza, J. (1975). From deviant to "normal": Factors associated with the official delabeling of criminally insane patients. Unpublished doctoral dissertation, Case Western Reserve University, Cleveland, Ohio. (University Microfilms No. 75-19194, 526199) Cocozza, J., & Steadman, H. J. (1974). Some refinements in the measurement and prediction of dangerous behavior. American Journal of Psychiatry, 131, 1012-1014. Cocozza, J., & Steadman, H. J. (1976). The failure of psychiatric predictions of dangerousness: Clear and convincing evidence. Rutgers Law Review, 29, 1084-1101. Colorado Revised Statutes 1973, 16-8-101 to 16-8-122. Conrad, J. P. (1985). The dangerous and the endangered. Lexington, MA: Lexington. Cross v. Harris, 418 F.2d 1095 (1969).

PAGE 107

Dietz, P. E. (1985). Why the experts disagree: Variations in the psychiatric evaluation of criminal insanity. The Annals of the American Academy of Political and Social Science, 477, 84-95. Dixon v. Commonwealth, 325 F. Supp. 966 [M. D. Pa. 1971]. Ennis, B. J., & Litwack, T. R. (1974). Psychiatry and the presumption of expertise: Flipping coins in the courtroom. California Law Review, 62, 693-7 52. Floud, J., & Young, W. (1982). Dangerousness and criminal justice. Totowa, NJ: Barnes & Noble Books. Gordon, R. A. (1977). A critique of the evaluation of Patuxent Institution with particular attention to the issue of dangerousness and recidivism. Bulletin of the Academy of Psychiatry and the Law, 210-255. Gordon, R. A. (1982). Preventive sentencing and the dangerous offender. The British Journal of Criminology, 22, 265-314. Greenland, C. (1965). Dangerousness, mental disorder, and politics. In C. D. Webster, M. H. Ben-Aron, & S. J. Hucker (Eds.), Dangerousness: Probability and Prediction, Psychiatry and Public Policy (pp. 25-40). New York: Cambridge University Press. Greenwood, P. W., & Turner, S. (March, 1987). Selective incapacitation revisited: Why the high-rate offenders are hard to predict. Santa Monica, CA: Rand for National Institute of Justice R-3397-NIJ. Hart, S.D., Kropp, P.R., & Hare, R. D. (1988). Performance of male psychopaths following conditional release from prison. Journal of Consulting and Clinical Psychology, 56, 227-232. Hermann, D. H. J. (1983). insanity acqui ttees: Automatic commitment and release of Constitutional dimensions. Rutgers Law Journal, 14, 667-662. Holland, T. R., Holt, N., & Beckett, G. E. (1962). Prediction of violent versus nonviolent recidivism from prior violent and nonviolent criminality. Journal of Abnormal Psychology, 91, 178-182. Holland, T. R., Levi, M., & Beckett, G. E. (1981). Associations between violent and nonviolent criminality: A canonical contingency table analysis, Multivariate Behavioral Research, 16, 237-242. Holland, T. R., & McGarvey, B. (1964). Crime seriousness progression, and Markov chains. Consulting and Clinical Psychology, 52, 637-640. 100 specialization, Journal of

PAGE 108

Iversen, G. R. (1984). Bayesian statistical inference. Beverly Hills: Sage. Janofsky, J., Spears, S., & Neubauer, D. (1988). Psychiatrists' accuracy in predicting violent behavior on an inpatient unit. Hospital and Community Psychiatry, 39, 1090-1094. Jones v. u. s., 103 set 3043 (1983). Katz, M. M., & Lyerly, S. B. (1963). Methods for measuring adjust-ment and social behavior in the community. Psychological Reports, 13, 503-535. Keilitz, I., & Fulton, J. P. (1984). The insanity defense and its alternatives: !!:. guide for policy makers. Williamsburg, VA: National Center for State Courts. Klassen, D., & O'Connor, W. A. (1988). A prospective study of predictors of violence in adult male mental health admissions. Law and Human Behavior, 12, 143-158. Klecka, W. R. (1980). Discriminant analysis. Beverly Hills: Sage. Koppin, M. K. (1977). A validation of the legal dangerousness scale with released criminally insane offenders. Paper presented at the Third National Conference of the American Psychology-Law Society, Snowmass, CO. Kozel, H. L. (1982). Dangerousness in society and law. The University of Toledo Law Review, 13, 241-267. Kozol, H. L., Boucher, R. J., & Garofalo, R. F. (1972). The diagnosis and treatment of dangerousness. Crime and Delinquency, 18, 371-392. Loeber, R., & Dishion, T. delinquency: A review. (1983). Early predictors of Psychological Bulletin, 94, 68-99. male Low, P. W., Jeffries, J.C., Jr., & Bonnie, R. J. (1986). The trial of John W. Hinckley, Jr.: A case study in the insanity defense. Mineola, NY: The Foundation Press. Margulies, P. (1984). The pandemonium between the mad and the bad: Procedures for the commitment and release of insanity acquittees after Jones v. United States. Rutgers Law Review, 36, 793-836. Martin, B. A. (1985). The reliability of psychiatric diagnosis. In C. D. Webster, M. H. Ben-Aron, & S. J. Hucker (Eds.), Dangerousness: Probability and prediction, psychiatry and public policy (pp. 65-86). New York: Cambridge University Press. 101

PAGE 109

Meehl, P. (1954). Clinical versus statistical prediction: A theoretical analysis and review of the evidence. Minneapolis: University of Minnesota Press. Megargee, E. I. (1976). The prediction of dangerous behavior. Criminal Justice and Behavior, 3-21. Megargee, E. I. (1981). Methodological problems in the prediction of violence. In J. R. Hays, T. K. Roberts, & K. S. Solway (Eds.), Violence and the violent individual. New York: Spectrum. Meloy, J. R. (1987). The prediction of violence in outpatient psychotherapy. American Journal of Psychotherapy, 41, 38-45. Meloy, J. R. (1988). The psychopathic mind: Origins, dynamics, and treatment. Northvale, N.J.: Jason Aronson, Inc. Millard v. Cameron, 373 F.2d 468 (1966). Millard v. Harris, 406 F.2d 964 (1966). Miller, L. (1987). Neuropsychology of the aggressive psychopath: An integrative review. Aggressive Behavior, 13, 119-140. Monahan, J. (1973). Dangerous offenders: A critique of Kozol et al. Crime and Delinquency, 19, 554-555. Monahan J. (1981). Predicting violent behavior: An assessment of clinical techniques. Beverly Hills: Sage. Monahan, J. (1984). The prediction of violent behavior: Toward a second generation of theory and policy. America! Journal of Psychiatry, 141, 10-15. Morse, S. (1983, Dec.). Predicting future dangerousness. California Lawyer, 16-18. Morse, S. (1985). Retaining a modified insanity defense. In R. Moran (Ed.), Annals of the American Academy of Political and Social Science, 477, 137-147. Mulvey, E. P., & Lidz, C. W. (1985). Back to basics: A critical analysis of dangerousness research in a new legal environment. Law and Human Behavior, _2, 209-219. Overholser v. Russell, 283 F.2d 195 (1960). Pasewark, R. A., Jeffrey, R., & Bieber, S. (1987, Spring). Differentiating successful and unsuccessful insanity plea defendants in Colorado. Journal of Psychiatry Law, 55-71. 102

PAGE 110

Pasewark, R. A., Pantle, M. L., & Steadman, H. J. (1982). Insanity acqui ttees and felons: A control study of their detention and rearrest. American Journal of Psychiatry, 139, 892-897. Pasewark, R. A., & Seidenzahl, D. (1979). Opinions concerning the insanity plea and criminality among mental patients. Bulletin of the American Academy of Psychiatry and Law, ']_, 199-202. Pfohl, S. J. (1978). Predicting dangerousness: The social construction of psychiatric reality. Lexington, MA: D. C. Heath & Co. Pogrebin, M. R., & Poole, E. D. (1986). Parole decision making in Colorado. Journal of Criminal Justice, 14, 147-155. Polstein, J. (1985). Throwing away the key: Due process rights of insanity acquitees in Jones v United States. American University Law Review, 34, 521. Pruesse, M., & Quinsey, V. L. (1977). The dangerousness of patients released from maximum security: A replication. Journal of Psychiatry and the Law, 293-299. Quinsey, V. L. (1979). Assessment of the dangerousness of mental patients held in maximum security. International Journal of Law and Psychiatry, 389-406. Reisner, R. (1985). Law and the mental health system: Civil and criminal aspects. St. Paul: West. Rolland, S. M. (1980). Recidivism among treated criminal psychiatric patients. Bulletin of the American Academy of Psychiatry and the Law, _, 15-27. Rosenberg, R., Knight, R. Prentky, R., & Lee, A. (1988). Validating the components of a taxonomic system for rapists: A path analytic approach. Bulletin of the American Academy of Psychiatry and the Law, 16, 169-185. Rubin, B. (1972). Prediction of dangerousness in mentally ill criminals. Archives of General Psychiatry, 27, 397-407. Schmidt, W. C. (1984, Winter). Supreme court decision making on insanity acquittees does not depend on research conducted by the behavioral science community: Jones v United States. Journal of Psychiatry and Law, 507-525. Segal, S., Watson, M., Goldfinger, S., & Averbuck, D. (1986). Civil commitment in the psychiatric emergency room: The assessment of dangerousness by emergency room clinicians. Archives of General Psychiatry, 45, 748-763. 103

PAGE 111

Shah, S. A. (1975). Dangerousness and civil commitment of the mentally ill: Some public policy considerations. American Journal of Psychiatry, 132, 501-505. Shah, S. A. (1978). Dangerousness: issues in law and psychology. 238. A paradigm for exploring some American Psychologist, 33, 224Shah, S. A. (1981). Legal and mental health system interactions: Major developments and research needs. International Journal of Law and Psychiatry, 219-270. Shah, S. A. (Ed.). (1986, March). The law and mental health: Research and policy. The Annals of the American Academy of Political and Social Science, 484. Silver, S. B. (1982). Testimony presented to the U.S. Senate in Limiting the insanity defense: Hearings before the Subcommittee on Criminal Law of the Committee on the Judiciary, United States Senate, Ninety-Seventh Congress (Serial No. J-97-122). Washington, D. C.: U.S. Government Printing Office. Singer, R. (1985). The aftermath of an insanity acquittal: The supreme court's recent decision in Jones v. United States, Annals of the American Academy of Political and Social Science, 477, 114-124. Spodak, M., Silver, S., & Wright, C. (1984). Criminality of discharged insanity acqui ttees: Fifteen year experience in Maryland reviewed. Bulletin of American Academy Psychiatry and Law, 12, 373-382. State v. Krol, 344 A.2d 289 (1975). Steadman, H. J. (1972). The psychiatrist as a conservative agent of social control. Social Problems, 20, 263-271. Steadman, H. J. (1980). The right not to be a false positive: Problems in the application of the dangerousness standard. Psychiatric Quarterly, 52, 84-99. Steadman, H. J. (1981). Special problems in the prediction of violence among the mentally ill. In J. R. Hayes, T. K. Roberts, & K. S. Solway (Eds.), Violence and the violent individual. New York: Spectrum. Steadman, H. J., & Cocozza, J. (1974). Careers of the criminally insane: Excessive social control of deviance. Lexington, MA: D. C. Heath & Co. 104

PAGE 112

Steadman, H. J., Cocozza, J., & Melick, M. E. (1978). Explaining the increased arrest rate among mental patients: The changing clientele of state hospitals. American Journal of Psychiatry, 135, 816-820. Steadman, H. J., & Keveles, G. (1972). The community adjustment and criminal activity of the Baxtrom patients: 1966-1970. American Journal of Psychiatry, 129, 304-310. Steadman, H. J., & Morrissey, J. P. (1981). The statistical prediction of violent behavior: Measuring the costs of a public protectionist versus a civil libertarian model. Law and Human Behavior, 263-274. In re Stephens, No. 77C-7229, Div. C. Adams County District Court, Colorado, July 23, 1987. Stone, A. A. (1982). The insanity defense on trial. Hospital and Community Psychiatry, 33, 636-640. Stone, A. A. (1985). The new legal standard of dangerousness: Fair in theory, unfair in practice. In C. D. Webster, M. H. BenAron, & S. J. Hucker (Eds.), Dangerousness: Probability and prediction, psychiatry and public policy (pp. 13-24). New York: Cambridge University Press. Thornberry, T., & Jacoby, J. (1979). community followup of mentally University of Chicago Press. The criminally insane: A ill offenders. Chicago: VandeCreek, L., Knapp, S., & Herzog, C. (1987). Malpractice risks in the treatment of dangerous patients. Psychotherapy, 24, 145-153. Vispo, R. H. (Ed.). (1980). The dangerous patient [Special issue]. Psychiatric Quarterly, 52. Webster, C. D., Ben-Aron, M. H., & Hucker, S. J. (Eds.). (1985). Dangerousness: Probability and prediction, psychiatry and public policy. New York: Cambridge University Press. Wenk, E. A., Robinson, J. 0., & Smith, G. W. (1972). Can violence be predicted? Crime and Delinquency, 16, 393-402. Werner, P. D., Rose, T. L., & Yesavage, J. A. (1983). Reliability, accuracy, and decision-making strategy in clinical predictions of imminent dangerousness. Journal of Consulting and Clinical Psychology, 51, 815-825. Wexler, D. B. (1983). The structure of civil commitment: Patterns, pressures, and interactions in mental health legislation. Law and Human Behavior, 7__, 1-18. 105

PAGE 113

Yesavage, J. A. (1984). Correlates of dangerous behavior by schizophrenics in hospital. Journal of Psychiatric Research, 18, 225231. Yodofsky, S., Silver, J.M., Jackson, W., Endicott, J., & lnlliams, D. (1986). The overt aggression scale for the objective rating of verbal and physical aggression. American Journal of Psychiatry, 143, 35-39. 106

PAGE 114

MEMORANDUM COLORADO STATE HOSPITAL Department of Institutions Division of Mental Health 1600 West Twenty-Fourth Street Pueblo, Colorado 81003 Phone (3031 546-4000 TO: Clinical Staff Preparing Information for Forensic Release Examinations FROM: David W. Rose, Ph.D., Chair of Forensic Disposition Committee (FDC) 2602 DATE: October 22, 1986 SUBJECT: Current Disposition Procedures I. Introduction Although the Disposition Committee has undergone considerable change recently, the change is seen as part of an ongoing development of procedures for insuring the orderly, and accurate, determination of eligibility for release of NGRI and NGRIMC patients. This function has been fulfilled by a committee under various names using various structures and procedures for nearly 25 years. This memorandum is to detail the present conception and operation of the committee. Separate memoranda will deal more specifically with clinical content and criteria for release. II. Purpose The purpose of the committee is to advise the Superintendent ofthe Colorado State Hospital concerning the release or of patients whose legal basis for enrollment is Not Guilty by Reason of Insanity or Not Guilty by Reason of Impaired Mental Condition. The advice to the Superintendent is in the form of approving or disapproving a psychiatric team recommendation. The committee does not claim original or superior knowledge of the patient even when the patient is interviewed, but rather, reviews available materials for internal consistency and. compatibility of rationales and conclusions with accepted policy, clinical knowledge and judgment. In its primary mission. of protection of public safety, committee judgments are expected to tend to be conservative more often than liberal. III. Composition The present membership of the committee includes a Chairman. appointed jointly by the Director of the Institute for Forensic Psychiatry and the Superintendent of the Colorado State Hospital. Other regular members of the committee currently include a Licensed Staff Psychiatrist, Assistant Director for the Institute for Forensic Psychiatry, Division _Chief Nurse, and the Director

PAGE 115

Clinical Staff -2-October 22, 1986 of Forensic Aftercare. In addition, Geriatric Treatment Center and .General Adult Psychiatry are each requested to provide a member on an alternating basis, and. the CSH Drug Treatment Program may be asked to provide an expert when the committee reviews a patient with a serious substance abuse problem. Each regular member is expected to identify at least one alternate who may be used in the absence of the regular member. The alternates should be able to provide the same kind of expertise and orientation as the regular members for whom they substitute. In addition to the voting membership outlined above, representatives of the patient's treatment team are encouraged to attend and to participate in discussion, but do not vote. Observers who are present to gain an understanding of the operation of the committee may attend at the discretion of the Chair but ordinarily do not participate in the discussion and do not vote. Patients' attorneys may attend interviews of their clients but attend discussions by the committee only upon invitation from the whole committee. The committee reserves the right to exclude all. observers and treatment team members from the meeting prior to the final vote. IV. Scheduling Patients are scheduled for Disposition Committee review when a court order for release examination is received by the hospital or when the patient's treatment team requests the patient be scheduled for review of a team recommendation for release. Scheduling is done through the Disposition Committee secretary who immediately informs the necessary individuals of materials that .. should be provided to the committee. Committee members are notified one week prior to a review of the patients who will be reviewed at that time. The secretary is responsible for maintaining adequate target dates to insure that materials can arrive in court at the required date. If it is clear that deadlines cannot be met, the Disposition Committee secretary must be informed so she may initiate a request for an extension from the court. Extensions are not requested lightly. The statute gives us 30 days for a release exam from the date of the order. Typically, the order is a week old by the time we receive it. That leaves three weeks for preparation and review of all reports. Upon receipt of the order, the FDC review of team materials will be scheduled for the Tuesday nearest two weeks away, the team two weeks and the committee and Administration. one week. In the case of team-initiated recommendations for release, team materials should be delivered to the Disposition secretary one week prior to the review date to allow study by the committee members before the meeting time.

PAGE 116

Clinical Staff -3-October 22, 1986 V. -.Preparation: Materials to be Submitted Materials submitted to the Superintendent and to the court are used to document and to explain decisions of great importance to the patient, to the hospital, and to the public. Materials are requested to provide an accurate and clear understanding of the case, and the factual basis for conclusions. To have adequate preparation time, the team should consider initiating evaluation when they first learn that the patient has petitioned for release. A .. A. comprehensive Psychosocial Chronology or history should be prepared under the direction of the team social worker. This chronology should focus more on detailed factual data in a chronological sequence than on conclusions or recommendations and is intended to serve as a detailed reference source, for the committee, members of the legal profession, future treatment teams, and aftercare workers in the event of the patient's release. As such, the original document source for important or controversial pieces of information should be cited. While the format is less important than the accuracy and thoroughness of content, the following general format is suggested. 1. Psychosocial history covering chronologically all aspects of the person's birth, childhood and adolescent development up to the point of the instant offense. Where criminal history is extensive or of particular note, it may be presented as a separate section, but there should be sufficient date-or age-related materials so the reader can understand how criminal activities fit into sequence with other important events in the patient's life. 2. Instant offenses (crimes leading to current hospitalization) should be described in considerable detail based primarily on official records, e.g. police and district attorney investigation reports, rather than on the patient's version or reports of professional persons who have primarily depended upon the patient's version. If the patient's version is significantly different from the official version, the nature of the differences should be pointed out. A brief synopsis of the expert opinions concerning a plea of NGRI or NGRIMC should be presented, as well as description of the patient's behavior before, during and after his offense.

PAGE 117

clinical Staff -4October 22 ,. 1986 3. The patient's hospital course should be presented from the date of admission to the date of writing the report. Hospital course should show the major trends both in terms of problems and acting out as well as significant situational changes and therapeutic advancements. Certainly, significant events in a patient's life, e.g., loss of a significant other, or significant financial change while a patient should also be noted. It is not usually necessary to cite each ward change as the committee will have a computer report of movement history for reference. 4. A detailed review of the last six months, including antisocial, psychotic, or other negative indicators should be available in as concise a form as possible, as well as a brief description of the patient's daily behavior at the time of updating the history. Any of the above sections may be incorporated into the Psychiatric Evaluation by team decision. If a single comprehensive history already exists in the chart, updating of that history may be accomplished with an addendum that includes correction of any misinformation and events that have occurred since that history was written. The addendum must clearly specify, by date and author, the comprehensive referred to. If a complete history is available only by surveying several sources, a new comprehensive history should be prepared, a procedure that will be much easier when computerized information services for that purpose are available. It is often found in the histories of our patients that there are serious discrepancies and gaps of information. When these are discovered, they should be highlighted rather than suppressed so the reader can be aware of the relative reliability of the information involved. B. Psychiatric Evaluation: The minimum statutory requirements for release examinations are listed in 16-8-106, C.R.S., as amended. Please also see the memorandum by Dr. Howard Fisher and myself, dated October 22, 1986, providing detailed considerations for preparation of the report. If the History or Psychosocial Chronology is available prior to initiation of the Psychiatric Examination, it may be referred to, saving time for the psychiatrist to focus on significant issues of formulation and clear documentation of behaviors relating to the evaluation of the patient's dangerousness. While an understanding of the patient and the treatment process -is important, the evidence and rationale for any conclusions concerning the patient's potential for dangerous behavior will be examined closely, not only by the committee but

PAGE 118

Clinical Staff -5October 22, 1986 by the Superintendent and by the courts. Frequently, the focus on difficult diagnostic or treatment issues or enthusiasm about progress shown_ by the patient leads examiners to present the objective information concerning potential danger in such a. way that the report fails to provide a sufficiently sound evidentiary basis for the conclusions or the conclusions are misunderstood in court. The basis and soundness of the conclusions offered in psychiatric examinations frequently become the primary issues in court. C. A recent Psychological Examination is always desirable, and is specifically requested if the patient is listed as -"Level II," .implying .greater than usual presumption of risk. Where possible, the psychologist's report should make reference to any trend or changes apparent in a sequence of testing over time. D. The Disposition Referral Form is intended to provide ready reference to significant data that may be needed by the committee or in preparing a report. It includes space for the team's opinions about the patient as well. E. All clinical records are to be brought to the meeting. This includes the current ward chart and all supplemental charts. Selected materials from the current chart will be duplicated so that it may be returned to the ward at the close of the meeting. Supplemental charts will be retained for use by the member who prepares the report, and will ordinarily be retained until the report. is.completed and signed. Case materials are expected to include at least the following Forensic materials: a copy of the order committing the patient as NGRI or NGRIMC; if there is more than one commitment, a copy of each commitment order should be available; FBI and/or CBI arrest records; if the patient has escaped for one week or longer, there should be an updated rap sheet covering the period of time during the escape even if there is every indication that the patient violated no law during that time; reports of police and/or district attorney investigations into the instant offense and any offenses alleged to have been committed during hospitalization. Instant offense reports should be sufficiently complete to allow staff to determine as nearly as possible just what occurred and, if possible, how. it occurred. If there is more than one instant offense, there should be similar information about each of the offenses. Finally, there should be copies of each observation report relating to the current commitment. That may involve competency observations as well as sanity

PAGE 119

Clinical Staff -6-'October 22, 1986 observations, and frequently involves reports written by experts outside the hospital and which are not automatically forwarded to the hospital. They are sometimes referred to in the court order committing the patient or in court orders requesting the report. Those reports are particularly useful in explaining why the patient was found NGRI or NGRI1>1C. Oftentimes, behavior has changed considerably by the time of the Disposition Cornmi ttee revie\-1, and it is difficult to track the nature of the changes that have taken place. F. The patient should be brought to the meeting for interview if there is a recommendation for release. In exceptional cases, with agreement bythe Chair, there may be determination of a need for the committee to interview a patient who has not been recommended for release. G. It is expected that whether the patient is interviewed or not, a team will be present to answer questions, particularly concerning current behavior of the patient. It is strongly recommended that the person present be a team member who is very familiar with the patient 1 s history, current behavior and treatment. If the patient has recently moved, the previous team may be asked to help. VI. Procedures in Meeting A. With team-initiated release recommendations, committee members will have read the submitted materials prior to the meeting. Otherwise, the. submitted materials, particularly the history and psychiatric examination report, are read aloud. B. There is discussion with the team representative and among the committee, particularly with a view towards identifying crucial issues and uncertainties in the case. C. If the patient is interviewed, there is an attempt to explore concerns, common to virtually every case, and further to explore issues that are apparent for this case in particular. There is an active effort to discover issues that have not been resolved or that have not been addressed in the submitted materials. It is expected that all significant issues will have been appropriately addressed in reports. The committee assumes no limitation on the areas that may be explored with the patient or on the depth to which these areas may be probed. It is the committee 1 s responsibility to satisfy itself that recommendations are well-grounded in tangible evidence and accepted clinical conclusions pertinent to the safety of society as well as the welfare of the patient.

PAGE 120

Clinical Staff -7-October 22, 1986 Interviews may be tape-recorded only with the patient's explicit permission. Marked anxiety about the interview is common and recognized as "normal" if appropriately acknowledged and managed by the patient. D. Visitors may be excluded from part or all of the discussion because of the necessity of providing an atmosphere in which hypotheses and observations may be explored freely within the committee. E. The committee will vote to approve or to disapprove a recommendation. If a recommendation favors conditional release, discussion should review major protective conditions to be applied. VII. Committee Reports A. The patient is notified by memo as to the actions of the committee and whether a second examination is expected. B. The author of the Disposition Committee report, who is ordinarily the Chair, reviews as much material as necessary to gain a coherent understanding of the case and to answer questions not answered during the meeting. Recent progress. notes, TPRs, correspondence, problem lists, and prior psychological and psychiatric reports may be ,reviewed. The intent of this review is to assure that contrary information is fully recognized, and that the important gaps in information are identified. The report usually will have the following sections: an identification section providing. quick reference to important case data; an administrative orientation or background section, clarifying unusual or complex features of the case from an administrative point of view and providing significant information not included in submitted reports. This section also tells the reader where other information is to be found. If the patient has been interviewed, one section summarizes the major topics discussed with the patient. Finally, .a discussion section should detail as well as possible the committee's understanding of the case through the information presented. The discussion section should lead primarily toward clear presentation of the reasons for the committee's decision, including recognition of the basis for any dissent within the committee. If the patient is felt to have an abnormal mental condition, nature of that mental condition should be specified as concisely as possible, and if the patient is felt to be dangerous to himself, others or to the community, the. nature of the expected dangerous behavior should be specified as clearly as possible.

PAGE 121

Clinical Staff -8-October 22, 1986 The purpose of the report is not to "sell" the committee s to the Supe::-intendent or to the courts. Clinicians have an obligation to spell out as clearly as possible what behavior is expected and under what circumstances. Often, it is not possible to predict future behavior with any strong assurance, and such limitations and known contradictory information should be recognized as well as that which supports the committee s judgment. It is then the task of the court in representing society. to make the social and philosophical judgment as to what kinds of behavior and risks are acceptable to society and what kinds of behavior merit continued removal of a person from society. Although the court procedures are constructed on an adversarial basis and it is the job of attorneys on both sides to "win" the case, it is our clinical responsibility to present information facilitating the court's decision-making process. Administrative and the review that are should be separately treatment observations during not pertinent to the court s decision directed toward appropriate admini-strators. C. Conditions: If the committee has approved a recommendation for release, the report should include a formal list of carefully worded conditions to accompany the hospital recommendation for conditional release. D. Second Opinions: Although the Superintendent may request any additional information or opinions .desired, a second expert opinion isrequired when release is recommended for a patient whose instant offense was a major act of potentially life-threatening violence (murder, kidnapping, rape, aggravated assault, and major cases of arson), or where the committee disapproves of the team or psychiatric recommendation. It is the responsibility of the Chair to request a second opinion, identifying the psychiatrist or psychologist who is to provide that opinion, and citing the reason for the request in a memorandum to that expert. Where specific points of disagreement are spelled out, the second examination may focus mainly on those issues. Any sources of informatiol) may be used. VIII. Transmittal of Documents In scheduling the review, a clear timetable is established with deadlines for each step of the process. All typed materials, once proofread and signed, are forwarded through the physician who participated in the committee review and the Director of the patient's treatment divisionwhb sign the report to acknowledge its content. It is then submitted to the Superintendent. The

PAGE 122

Clinical Staff -9CSH Legal Administrator reviews of proper legal presentation. The and responsibility in making. a Any additional information may be to aid in making a decision. October 22, 1986 the materials for assurance Superintendent has full freedom recommendation to the court. requested by the Superintendent Only the Superintendent's recommendation may be taken as the hospital's recommendation or official position. Any alternative view presented by an employee must be presented as one' s own personal professional opinion with the hospital's position clearly acknowledged. The Superintendent is, of course, free to alter suggested conditions for a condi tiona! release in any way that appears appropriate. IX. Court Action: If no expert in or out of the hospital favors conditional release or discharge, a reiease hearing will not be held. The court has final authority upon hearing all evidence presented and is not legally bound by the recommendation of any participant. Occasionally, patients become quite anxious upon hearing that a team or the Disposition Committee has recom-. mended against their release. They should understand that the court is the final arbiter and will be considering information presented by the patient s attorney as well as that presented by the hospital. X. Clinical Management of Patients in the Process A. It is inappropriate to coerce the patient into withdrawing or altering a petition for release. The clinical implications may be discussed with the patient but the choice of whether to file a petition or proceed with one, no matter how hopeless it may seem to be, is strictly up to the patient and his attorney. B. "Meeting Disposition" is one of the most anxiety-provoking hospital events for many patients. The team should offer such additional support as realistically indicated, and reassure the patient that moderate anxiety about the review, if appropriately managed, is not a primary consideration concerning release . C. Many patients who are seeking release via a petition put treatment "on hold," pending the outcome of the petition. While efforts should be made to maintain the patient in treatment during this time, and often a wise defense attorney will urge his client to be cooperative in order to get a better report, the patient's attitude and behavior during this time period may not be predictive of his adjustment

PAGE 123

Clinical Staff -10-October 22, 1986 DWR:lc following a court decision. Personnel should anticipate that as time approaches for Disposition Committee review and for a court hearing, the patient's anxiety level may increase considerably. Some patients are inclined to act out in order to assure their non-release. If the decision goes against the patient's wishes, the patient should be evaluated for a possible increase in escape risk until he adjusts to that result.

PAGE 124

!602 MEMORI\NDUM COLORADO STATE HOSPITAL Department of Institutions Division of Mental Health 1600 West TwentvFourth Street Pueblo, Colorado 81003 Phone (303) 546-4000 TO: FROM: DATE: SUBJECT: CSH Psychiatrists Howard W. M.D. and David W. Rose, Ph.D. Forensic Disposition Committee October 22, 1986 Crucial Information in Psychiatric Reports Required by the Disposition Committee_to Make Adequate Assessment This memorandum is supplementary to the broader memo entitled "Current Disposition Procedures," dated October 22, 1986. See also "examination and report," C.R. S. 16-8-106, as amended. To the standard format of psychiatric evaluation involving a standard medical history and mental status examination, there should be specific discussion in the following areas: 1) Last known dangerous behaviors 2) Last known illegal behaviors 3) Last known suicidal ideation 4) Discussion of the instant offense -comparing the patient's version with the police reports. This should be done in two separate paragraphs. 5) Treatment formulation what exactly is being treated and why. Dangerousness is the paramount consideration, and dangerousness in the foreseeable future is what the Disposition Committee must address, by statute. 6) Why was the patient found Not Guilty by Reason of Insanity? Start from the original sanity examinations and work through the specific details of why and how the patient lost control of his behavior. Each of these formulations should have been addressed in the treatment plan formulation. How does each issue stand now? Which have been met in treatment and which have not been met in treatment? If there was a functional psychosis involved, why was it exacerbated? What stresses were operative? In what ways were the stresses resolved? If the illness

PAGE 125

has entered remission without medication, why is that so? What has been treated? What is the treatment team treating now? If the original disorder no longer exists, such as an organic condi-tion, what is the current diagnosis, and the treatment formulation? Since frequently we are dealing with elements of a personality disorder underlying other clinical entities, please be specific. How does this current treatment impinge on considerations of dangerousness? For example, if the basic element being treated is mistrust, or anger, or frustration and disappoint ment, what is being done .to the personality to result in successful handling of those areas of psychopathology? We must assess the 11Success11 of the treatment to date. There must be clear objectives. How do you know that the problem is resolved enough to discontinue treatment, or to move treatment to an outpatient setting? 7) Outpatient compliance -What problems do you ex pect this patient to have when he leaves the hospital? If there is a bistory of his being delusional, and that the symptom is only controlled with medication, how do you know he will be compliant as an outpatient? How does one know that the current medication is the medication of choice? It would be helpful to know which medications have not worked, .and which have worked, and whether, if the current works, there is a documentation of adequate serum level. Outpatient compliance with this medication is likely to be monitored, as in Lithium, with appropriate serum levels. Serum levels are particularly important where, even with medication, the patient is not in full remission, but there is. nevertheless a recommendation for conditional release. It should be specifically noted whether the medication has induced a full remission or not. Finally, if the patient is not seen as dangerous now, why is he not dangerous? Assumptions and evidence need to be presented clearly in order to communicate, both to the Disposition Committee and later to the court, why this patient can be considered safe outside the hospital. What will guard against decompensation? What will detract from compliance? What will stop this patient from committing the .very same offense at a later date, given the same circumstances that pertained when he committed the first instant offense? lf the chart indicates sever a 1 diagnoses by different physicians; the ,c_urrent physician should discuss adequate differential diagnosis, clarifying why one -.2

PAGE 126

particular diagnosis is considered the "correct" diagnosis. A psychiatrisf should realize that the report that he renders is going to have to be read by lay persons, including lawyers, judges, andjurors .It is particularly that there be clear understanding of the particular statute being addressed because release criteria are different, de pending upon the date of the patients instant offense. 16-8-120 states that instant offenses occurring prior to June 2, 1965 involves whether the patient is still sane or insane. Since 1965, until June 30, 1983, the test for release from commitment, or eligibility fbr conditional release, is no abnormal mental condition that would likely cause dangerousness in the foreseeable future." since. July the test has been embellished to indicate no abnormal mental condition causing dangerousness iri the reasonably foreseeable future, and the patient is capable of distinguishing right from wrong has substantial capacity to conform his conduct to the requirements of law. This additional qualifier is not part of the IMC test.for release. There has apparently been no judicial test as to whether the above language constitutes a criteria for release when the hospital is recommending release, as opposed to when the patient petitions for release. But, the patient may be released, according to statute, "by hospital authority, .. which is Dr. Haydee Kort. Your evaluation and recommendations, combined with Disposition Committee review, help Dr. Kart make appropriate decisions. The committee would welcome any questions that this notice generates. Please call Dr. David Rose, Chairman of the Disposition Committee,.or Dr. Howard Fisher, to discuss your ideas. Staff Psychiatrist DR:HWF:njc -3

PAGE 127

MEMORt\NDUM HOSPITAL Department of lnsttutions Division of Mental Health 1600 West TwentvFciurth" Pueblo, Colorado 81003 Phone (303) 546-4000 TO: FROM: DATE: Team Leaders and Disposition Committee Members Wnliam H. Ross, D1rector, IFP December 29, 1986 SUBJECT: Policy Regarding Release of Patients from Maximum and Medium Security Any recommendations to release criminally committed patients from maxi mum or medium security areas should only be made in excep tional cases. To justify this recommendation, there must be sufficient evidence to establish that it is not necessary or desirable to test the patient s demonstration of control (ability to function) in unsupervised situations. 2602 The individual examiner and the team must be responsible for assessing the potential dangerousness of a who will have increased freedom in society, and to be aware of the fact that this assessment is being made when the patient is confined in a restrictive area w-ith constant supervision. Such recommendations must be discussed with the Division Director, but only after an individual case review has been conducted. The ICR shall address questions of potentials for dangerousness. WHR:wa cc: Haydee Kort, M.D. Pat Robb David Rose, Ph.D.