University of Southern California Dissertations and Theses

The effects of framing and actuarial risk probabilties on involuntary civil commitment decisions

(USC Thesis Other)

The effects of framing and actuarial risk probabilties on involuntary civil commitment decisions

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content

THE EFFECTS OF FRAMING AND ACTUARIAL RISK PROBABILTIES

ON INVOLUNTARY CIVIL COMMITMENT DECISIONS

by

Nicholas Ivan Scurich

A Thesis Presented to the
FACULTY OF THE GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
MASTER OF ARTS
(PSYCHOLOGY)

December 2009

Copyright 2009 Nicholas Ivan Scurich

Acknowledgments

I would like to thank my advisor, Richard John, for all the help throughout the

years. I also wish to thank Thomas D. Lyon for giving me my first exposure to civil

commitment and the relevant issues. Additionally, to David J. Weiss for many thoughtful

discussions about experimental methodology.

December, 2009

ii
Table of Contents

Acknowledgments ii

List of Tables iv

List of Figures v

Abstract iv

Chapter 1. Introduction 1
The Status of Violence Risk Assessment 1
The Communication of Violence Risk 4
Current Aims 8

Chapter 2. Methods 15
Participants 15
Procedure and Design 15
Theoretical Design Considerations 16

Chapter 3. Experimental Analysis 19
Within-Subjects Analysis 19
Between-Subjects Analysis 22
Resolution 25

Chapter 4. Evaluation of Assumptions 28
Base Rate Assumptions and Putative Commitment Thresholds 28

Chapter 5. Discussion 32

Bibliography 39

Appendix 44
Case Example 45
iii

List of Tables

Table 1. Percentage of Within-subjects Commitment Decisions 19

Table 2. Percentage of Between-subjects Commitment Decisions 22

iv
List of Figures

Figure 1. Within-subjects proportion of decisions to commit by risk level 20
and probability frame

Figure 2. Within-subjects mean likelihood rating of decision to commit 21
by risk level and probability frame

Figure 3. Between-subjects proportion of decisions to commit by risk 23
level and probability frame

Figure 4. Between-subjects mean likelihood rating of decision to commit 24
by risk level and probability frame

Figure 5. Distribution of Base Rate Assumptions 30

v
Abstract

Research on how actuarial risk assessment results are to be communicated and
what effect they have on decision makers has received scarce empirical attention. Using a
2X3 factorial design, this experiment found that decision-makers are sensitive to
differences in actuarial risk probabilities, but that the decision for commitment largely
depends on how the risk probability is framed. A derivative of Support Theory (Tversky
& Koehler, 1994) is used to explain this violation of extensionality. Furthermore,
contrary to public opinion surveys, analyses of base rate assumptions indicate that
decisions makers do not have unrealistic perceptions of mental disorder and violence.
Substantive and theoretical implications are discussed.
vi
Chapter 1. Introduction
The Status of Violence Risk Assessment
Legal decisions to deprive life and liberty often hinge on the risk of violence
posed by individuals. In some states, for example, the imposition of capital punishment
requires a finding of dangerousness, loosely defined as posing a continuing threat of
violence (Barefoot v. Estelle, 1983). There has recently been a proliferation of statutes
that authorize the ex post confinement of dangerous sexual predators (Kansas v.
Hendricks, 1997), and all American states have statutes that authorize the involuntary
civil commitment of dangerous individuals with a mental disorder (O’Connor v.
Donaldson, 1975; Addington v. Texas, 1971). Thus the prediction of violent behavior
occupies an important role in the legal system. It is mental health professionals who are
frequently called upon to forecast violent behavior (Monahan and Steadman, 1994).
However, this practice is highly controversial because some dispute the ability of mental
health professionals to make predictions with sufficient accuracy (Melton, Petrila,
Poythress and Slobogin, 1987). Although initial objections appealed to professional and
ethical grounds, this controversy eventually made it to the Supreme Court, which
grappled with whether predictions of violence are sufficiently accurate to be legally
admissible as expert testimony.
When the issue came before the Supreme Court in the 1980’s (Barefoot v. Estelle,
1983), research on the validity of clinicians to predict violent behavior was fledging. The
Barefoot Court quoted the then current, now infamous result, “Psychiatrists and
psychologists are accurate in no more than one out of three predictions of violent
behavior (Monahan, 1981, p. 47).” It is this low rate of accuracy that led many
1
commentators, including the American Psychiatric Association (1983), to conclude such
predictions are not ‘expert’, or are at least outside the scope of mental health expertise.
However, the Supreme Court was not dissuaded by the research, and reaffirmed the legal
permissibility of psychiatrists and psychologists to predict violent behavior (Barefoot v.
Estelle, 1983).
Well before this ruling, Paul Meehl (1954) had been advancing the actuarial
approach to making general predictions. Unlike a clinical approach, which relies upon
subjective and impressionistic judgments, an actuarial approach uses empirically
established relations with explicit rules to render objective predictions (Dawes, Faust &
Meehl, 1989). Buttressed with empirical data, Meehl went on to argue that an actuarial
approach would almost always outperform unstructured clinical judgment. Grove and
Meehl (1996) outline a few reasons for this consistent superiority, which include that
clinical judgment is particularly susceptible to: ignoring base rates, assigning non-optimal
weights, failing to account for covariation, and perhaps most importantly, relying on
invalid cues (see pgs. 22-24). As replication began to accumulate across disciplines, to
date over 136 studies confirming the superiority of the actuarial approach (Grove et al.
2000), this slowly gave way to a paradigmatic shift in the research on violence risk
assessment. Instead of focusing on the predictive validity of clinical judgment, the
research field is now concerned with isolating empirically valid risk factors (Monahan,
2005). The implementation of empirically valid risk factors in risk assessment would
eventually take on many forms, most notably in ‘actuarial risk assessment tools’.
Actuarial tools remove clinical judgment in the prediction process because they
specify apriori what risk factors are to be used and how these factors are to be combined
2
or weighted to yield an objective estimate of risk (Underwood, 1979). In addition to
selecting the most predictive risk factors, as determined by empirical research, an
actuarial tool is able to optimally combine the risk factors to ensure the final estimate is
objective. In this approach, clinical judgment is only necessary to the extent it can
appraise risk factors required by the actuarial tool, otherwise it is largely eschewed. The
actuarial tool can be roughly thought of as the culmination of these processes in which an
algorithm is established for assessing violence risk.
Only in the past two decades have actuarial tools been developed specifically to
assess violence risk. The three most well-known risk assessment tools are: The Violence
Risk Appraisal Guide (VRAG; Harris et al. 1997); The Historical, Clinical, Risk 20
(HCR-20; Webster et al. 1997); The Classification of Violence Risk (COVR; Monahan et
al. 2005). These tools all yield predictive accuracy that exceeds unstructured clinical
judgment, and all produce a categorical or probabilistic estimate of risk. The COVR is the
most recent and yields the greatest predictive accuracy of any actuarial tool (Banks et al.
2004).
The COVR is the result of the Macarthur Study of Mental Disorder and Violence,
the most comprehensive study in the field to date (Monahan et al. 2001). The study
prospectively tracked more than 1,000 patients as they were discharged from psychiatric
facilities in three different states. The patients were assessed on over 134 risk factors and
were tracked in the community for twenty weeks following discharge. Patients were
monitored for violence by three different methods, which include self-report, official
records (i.e. arrest reports; re-hospitalization records, etc.), and collateral informants.
Forty factors emerged as significant predictors of violence. These factors were then
3
combined with a classification tree methodology, known as the “ICT” methodology,
which relies on a more interactive and contingent model of violence (Banks et al. 2004).
Unlike regular regression models, which require the same questions of all patients, the
ICT asks questions that are contingent on the previous response, so there is no pre-
established sequence of questions, and each patient would be asked a unique set of
questions based on their unique diagnostic composition (Monahan, 2008). This process is
done recursively until the patient is placed into one of five risk groups; each risk group
has an associated probability of violence: Risk Group 1=1%; Risk Group 2=8%; Risk
Group 3= 26%; Risk Group 4=56% and Risk Group 5=76%. The COVR is now available
as interactive software where risk probabilities are automatically computed.
The Communication of Violence Risk
The sole purpose of risk assessment is to inform legally empowered decision
makers (Heilbrun, 1997). The scope of risk assessment is confined to a descriptive nature
of information gathering, where the normative or legal issue of evaluation and action are
reserved for a separate decision maker (Schopp and Quatrocchi, 1995). This implies that
the assessment must be articulated properly if it is to be of any assistance to legal
decision makers. This realization engendered an area of research known as violence risk
communication.
The primary interest of violence risk communication is determining the most
efficacious and accessible way to communicate risk to decision makers (Schopp, 1996).
This issue is particularly pressing with actuarial tools because of their probabilistic
nature, which is known to be problematic for human decision-makers (Nisbett and Ross,
1980). The gains in predictive accuracy the actuarial approach affords might be useless,
4
or even worse than useless, if the results cannot be used meaningfully (Heilbrun,
Dvoskin, Hart and McNeil, 1999). Interestingly, two decades ago the National Research
Council (1989) identified risk communication as a broad area in need of study, and yet
the research on the communication of violence risk remains nascent.
Before any empirical endeavors were undertaken, a series of articles in American
Psychologist critically considered some theoretical issues underlying the apposite form of
violence risk communication. Monahan and Steadman (1996), drawing on an analogy
from weather forecasting, contended that assessment framed in terms of a probability is
most appropriate (contains the most relevant information), but that a categorical format
might be more efficacious for decision makers. Notwithstanding the empirical inquiry,
Schopp (1996) noted that categorical estimates of risk are more than a merely collapsed
and discrete form of probability. In the context of weather forecasting, categorical
estimates include prescriptive measures (i.e. you should evacuate) because the National
Weather Service doubts the ability of decision makers to comprehend and respond
appropriately to what is being articulated. Schopp (1996) argued this type of
prescription, either explicit or implicit, usurps the appropriate scope of risk assessment
because it involves value judgments reserved for legally empowered decision-makers.
Although it varies with situation, Schopp resolved the probabilistic format is most apt.
There was, nevertheless, unanimity that research was necessary to answer some of the
empirical concerns regarding the normatively correct form of risk communication.
Empirical investigations of violence risk communication have been scant; six as
of 2004 (Heilbrun et al. 2004), of which only three investigated the efficacy of the style
and content. These three studies will be briefly reviewed as the others generally concern
5
risk communication preferences. In the original empirical investigation of the topic,
Slovic and Monahan (1995) asked young adults and experienced forensic clinicians to
judge a) the probability that a hypothetical patient would commit violence; b) whether the
patient should be categorized as “dangerous”; and c) whether legal coercion should be
used to mandate treatment. In addition to finding that probability and “dangerous”
judgments were systematically related and could predict the decision for legal coercion,
they found that the response scale had an effect on the probability estimate of the patient
committing violence. Specifically, one response scale consisted of an 11-point scale
going from 0 to 100 in units of 10, and the other response scale included five small
probabilities (e.g. less than 1/1000, 1%, 2%, 5%, 10%, etc) and ended with “greater than
40%”. The latter scale leads to judgments of the patient being less dangerous and reduced
the call for legal coercion; this effect was the same for the young adults and experienced
forensic clinicians. The scaling effect suggests that the ordinal relation of risk
probabilities is useful but the estimated probability itself (magnitude of difference) is not.
It should be noted that generating a subjective risk probability estimate is unnecessary
with the application of an actuarial tool, because it is the tool that produces the
probabilistic estimate of risk.
Developments in cognitive psychology suggest people can aggregate and
communicate probabilistic information more effectively as a frequency (i.e. 1 out of 10)
than as a probability (i.e. 10%) (Gigerenzer & Hoffrage, 1995). In the second empirical
study Slovic, Monahan & Macgregor (2000) tested whether similar format effects would
be apparent in the communication of violence risk. In addition, they investigated whether
instruction on probabilistic reasoning could facilitate better risk judgments. They found
6
that instruction on the use of base rates and warnings about response scaling effects was
ineffective. Furthermore, consistent with Slovic and Monahan (1995), the response
alternatives systematically influenced the subjects’ answers. When the alternatives were
clustered around low probabilities (or low frequencies), even experienced clinicians
assumed the target behavior occurred infrequently. Their most interesting finding had to
do with the format of the response scale. The frequency format led to lower likelihood
estimates of harm than did the probability format; however, the frequency format also led
to higher perceptions of identical risk than did the probability format, an apparent
paradox. This result held irrespective of the relative frequency (e.g. 2 out of 10 or 20 out
of 100 vs. 20% probability). The authors attribute this paradox to subjects “imagining the
numerator”, where the numerator (e.g. 2 or 20) evokes images of harmful attacks or
violence. It is important to remember that since the actuarial tool produces the probability
estimate, the relevant finding here is that the frequency format effects the perception of
risk.
If it is easier to visualize frequencies than probabilities, and this is why subjects
perceive greater risk in the frequency format, Monahan et al. (2002) hypothesized that
other influences on visualization would affect perceived risk. The third empirical study
varied the format in which the risk was communicated (e.g. frequency vs. probability)
and vividness of the outcome description. That is, the subjects, who were experienced
clinical psychologists, were either given a pallid outcome depiction, where the violence
description was subtle, or a vivid outcome depiction where the description of violence
was salient. This study did not require the clinicians to generate a risk estimate, rather a
risk estimate generated from an actuarial tool was provided. When asking for a discharge
7
decision, the frequency format effect of Slovic, Monahan & Macgregor (2000) was
replicated only for clinicians who work in a forensic setting (the sample was
disaggregated into clinicians who work in a forensic setting and those who do not). For
clinicians who work in a forensic setting, this effect was more pronounced when the
description was vivid compared to pallid. The pallid description/ probability format
condition led to the least perceived risk, whereas the frequency/ vivid format led to the
greatest perception of risk. The authors attribute this to forensic clinicians being
“sensitized to the effects of fear-increasing risk communication formats (i.e. frequencies)
and fear-increasing depictions of the outcomes whose risk is being assessed (i.e.
vividness) (p. 126).”
Current Aims
Heilbrun et al. (1999) note: “The form of risk communication may…have the
potential to mislead, as when a court must determine whether an individual is ‘dangerous
to self or to others’ as part of a civil commitment proceeding. However, the accuracy
with which such information about risk is conveyed in such legally prescribed form has
not been studied or debated (p.95).” The current study investigates the proposition that
the form of risk communication- specifically the probabilistic form- is inherently
misleading by examining the coherence with which judges evaluate risk probabilities.
The study also systematically investigates whether decision makers are sensitive to the
effects of differential risk probabilities in making involuntary civil commitment
decisions. Both aims are elaborated upon below.

8
The effects of framing
The convention by which to frame communications of violence risk is as the
probability of being violent (P(V)). For example, individuals in Risk Group X have a Y%
probability of engaging in violent behavior. According to the axioms of probability
theory, any mutually exclusive proposition has to be either true or false, from which it
follows that the probability and its complement equal 1 (Baron, 2002). That is, if there is
a Y% probability the proposition is true, there is a 1-Y% probability the proposition is
false. In the context of risk assessment, for example, if individuals in Risk Group X have
a Y% probability of engaging in violence, individuals in Risk Group X also has a 1-Y%
of not engaging in violence. Although a coherent decision-maker would understand
these probabilities correspond to the judgment, empirical research suggests that human
decision-makers tend to violate this axiom.
A fundamental assumption of subjective probability theory is the principle of
extensionality, which roughly refers to the idea that events with the same extension
should be assigned the same probability. That is, the perceived probability of an event
should not change if the event is decomposed into its constituent components. Although
normatively correct, the principle is often violated because different descriptions of the
same event lead to vastly different judgments (Tversky & Koehler, 1994; Mulford and
Dawes, 1999; Sloman et al. 2004). For example, Tversky & Koehler (1994) discuss the
research of Johnson, Heresy, Meszaros and Kunreuther (1993) in which subjects were
presented with hypothetical health care that would either cover hospitalization for any
reason, or hospitalization for any type of accident or disease. Subjects who received the
latter option were willing to pay a higher health insurance premium, presumably because
9
they were more able to imagine being hospitalized and therefore thought the event was
more probable. Tversky and Koehler (1994) offer a reason for why extensionality fails
descriptively: “extensionality may fail because different descriptions of the same event
may call attention to different aspects of the outcome and thereby affect their relative
salience (p. 548).” This relative salience is consequential for probability judgments and
the evaluation of uncertain events (Rottenstreich and Tversky, 1997).
There is reason to suspect the framing of probabilities is consequential for
decisions involving civil commitment. For example, a possible explanation of reluctance
to discharge patients found in Monahan et al. (2002) study is that the description was
exclusively oriented towards the implications of erroneously releasing the patient (a false
negative); that is, the probability was framed in a way that emphasizes false negatives
(i.e. the probability of violence occurring). It is likely that since only one type of error
was emphasized, only that error was visualized; hence, clinicians were reluctant to
discharge patients because only the effects of false negatives were salient in their mind.
Would this reluctance hold if the other type of error (a false positive- needlessly
confining a patient) were made more salient? As previously noted, it is convention to
frame the probability in terms of being violent, but no one has yet to consider whether
this particular frame lends itself to inherent bias. Because decisions involving
preventative confinement are explicitly premised on balancing the costs of false positives
with the costs of false negatives (Monahan and Wexler, 1978), the current experiment
investigates whether the framing of probability leads to an inherent bias towards false
positives or false negatives, and if this bias affects civil commitment decisions.
10
It should also be noted that the general reluctance could also be a corollary of
clinicians practicing “conservative medicine”, where a false positive may be of minimal
consequence to the clinician- even if it is serious for the patient- compared to the drastic
consequences to the clinician for a false negative. This raises a more general concern
with the existing empirical research, which is that it has inappropriately used clinicians as
subjects, thereby failing to examine the effects of risk communication on actual
consumers of risk assessment. The concern is not that clinicians are inept consumers,
rather the concern stems from the fact the function of the clinician and the consumer
differs in principle (Schopp and Quattrocchi, 1995). The clinician is responsible for
assessing and communicating risk to a “consumer” who is responsible for making an
evaluative decision (i.e. a commitment decision); the clinician expresses risk, the
consumer evaluates risk. Unlike a clinician who would consider professional liability,
reputation and ethical obligations in making evaluative decisions (akin to those in
Monahan et al. 2002), a “consumer”, in theory, is a disinterested party appointed to weigh
competing interests and make legal determinations objectively. Thus the generalizability
of the prior empirical research is questionable because the roles and interests of clinicians
are confounded. The current experiment disentangles these roles by focusing on how
consumers evaluate actuarial risk assessment.
The effects of differential risk probabilities
A major assumption of actuarial risk assessment is that decision-makers are
sensitive to differences in risk probabilities
1
. That is, it assumed that decision makers can

11
1
I shall use the locution “risk probability” to refer to the probability of violence occurring. A probability is
typically subsumed under the concept of risk, which includes other features such as the magnitude of the
distinguish between different probabilistic risk levels, and that varying risk probabilities
are consequential for legal decisions. However, no study has adequately tested this
assumption. As noted previously, the benefits of the actuarial approach might be negated
if this assumption is false.
One recent study, which was concerned with retributivism and the commitment of
sexually violent predators, included different risk probabilities (Carlsmith, Monahan and
Evans, 2007). The study used extreme risk probabilities (i.e. 0%, 4%, 70% probability of
recidivism), which theoretically should have straightforward implications for decisions.
However, the researchers found the relation between elevated risk probabilities and
commitment decisions to be non-monotonic. Subjects, who were both students and jury
eligible citizens, endorsed commitment more when then there is a 0% probability than
when there is a 4% probability of recidivism. This result questions the underlying
assumption of risk assessment which is that decision makers are sensitive to differences
in risk probabilities. Moreover, the difference between the 0%/4% probabilities and 70%
probability of recidivism, albeit statistically significant, was not as pronounced as one
would expect. A conservative interpretation of these results suggests the probability of
risk has an insignificant effect on legal decisions.
Surprisingly, only one study has examined the isolated effects of actuarial risk
data on the decisions rendered by actual judges (Monahan and Silver, 2003). In the study,
judges were asked to indicate what level of risk satisfies the legal threshold for
authorizing involuntary civil commitment, based on the risk probabilities associated with
the five different risk groups from the COVR. There was significant variability as to the

outcome (Vose, 2000), however I use the term risk probability strictly as the probability of occurrence
12
legal threshold for commitment; most judges claimed somewhere between 8% and 26%
was sufficient risk to authorize involuntary commitment and three judges considered 1%
sufficient. Without a long excurses, there is legal reason to believe some of the variability
in the putative thresholds is illegitimate (Scurich and John, 2009), particularly for the 1%
threshold since civil commitment is predicated on the assumption not everyone with a
mental disorder is legally dangerous (Monahan, 2006). The variability in the putative
thresholds could represent an idiosyncratic cost-utility function of each judge, or it could
be that judges had difficulty reconciling probabilities with legal concepts. Although this
study did not examine the effects of different risk probabilities on commitment decisions,
or if the putative thresholds would actually be applied to a real case (i.e. if presented with
a case in which the risk probability was less or more than the putative threshold, would
the judge release/commit, respectively), the results underscore the general concern of
Heilbrun et al. (1999) that research on better-informed legal decision making is
underdeveloped.
The current study investigates whether or not judges are sensitive to different risk
probabilities by having judges make civil commitment decisions in cases that differ in
their relative risk. In so doing, this approach demonstrates what is actually done, not what
judges claim they would do. Furthermore, this study does not only use risk groups
‘extreme’ risk probabilities; it is the indeterminate risk groups that force decision makers
to critically evaluate the relative costs.

without consideration of magnitude.
13
Specific hypotheses
I hypothesize a main effect for framing with an ordinal interaction for the
moderate risk level where the effects of framing are likely to be most pronounced. In
particular, I hypothesize that medium risk framed as violent is sufficient to justify civil
commitment, but medium risk framed as non-violent is insufficient to justify civil
commitment.

14
Chapter 2. Methods
Participants
303 university undergraduate students participated as mock judges in this
experiment. The sample had a sex distribution of 70 males and 233 females. The age
range was 18 to 37 with a mean 20.46 (S.D. = 2.42) and median 20.00 (IQR = 2). A
minority of the subjects (27%) indicated they personally knew someone who has been
committed to a psychiatric facility. 19% of subjects consider themselves conservative,
46.5% consider themselves moderate and 34.5% consider themselves liberal. Subjects
received partial course credit for their participation.
Procedure and Design
Subjects completed an anonymous online questionnaire. The first page included a
brief summary of what the experiment entailed, as well as a disclosure of rights that
indicated subjects may decide to stop participating at any point during the study without
penalty. After agreeing to continue, subjects then read six individual case summaries and
rendered judgment on each. The case summaries were adapted from Monahan (2008) and
are “case examples of the use of the COVR in clinical practice (p. 25).” The case
summary included a dangerousness risk estimate and its corresponding range, as well as
additional information about the respondent, including his age, psychiatric diagnosis and
reason for the current referral to the psychiatric facility. The primary dependent measure
was a dichotomous decision to either authorize a 14-day involuntary psychiatric
commitment (i.e. ‘commit’) or to not authorize the involuntary commitment (i.e.
‘release’). In addition, subjects also rated how likely they were to either commit or release
on a 6-point scale. The scale ranged from 1 “very likely to release” to 6 “very likely to
15
commit” where in-between 3 and 4 was a transition point from release to commit,
respectively. Ambivalence was not permitted and subjects had to a make determination
either way, as is the case in juridical practice. After disposing of all six cases, the subjects
then provided demographic information, as well as political views and general views
regarding mental disorder and violence.
This experiment employed a 2 (framing: probability of violence [P(V))] or
probability of non-violence [P(1-V)]) X 3 (risk level: low, i.e. 1%; moderate, i.e. 26%; or
high, i.e. 76%) Latin Square factorial design. This yielded six different vignettes that
reflected six different hypothetical cases. For example, one vignette contained a moderate
probability of risk framed as violent, thus subjects were told, “there is a 26% probability
of violence occurring.” Another vignette also contained a moderate probability of risk
framed as non-violent, thus subjects were told, “there is a 74% probability of violence
not-occurring.” Each estimated probability contained a confidence interval with
reassurances that a clinical interview failed to uncover any countervailing that would call
into question the risk estimate. A full example of the vignette can be found in the
Appendix.
Theoretical Design Considerations
Shortcomings of a between-subjects design are well known within the social
sciences (Fisher, 1925). In addition to having a smaller sample size, which has obvious
implications for statistical power (Cohen, 1977), between-subjects designs heavily rely
on random assignment to equally distribute error variance amongst subjects in different
conditions (Anderson, 1981), and typically this assumption goes unsubstantiated
(Birnbaum, 1999). Part of the unequal error variance found in between-subjects designs
16
can be explained individuals imagining different contexts when making judgments. For
example, Birnbaum (1999) demonstrated with a between-subjects design that subjects
judged nine to be significantly larger than 221 based on contextualizing the range of
possible numbers (i.e. a subject was either asked to judge “how large is the number 9” or
“how large is the number 221”? The latter condition is claimed to have invoked a context
of one-digit numbers, hence 9 is a “large number”, whereas the latter evoked a context of
three-digit numbers, hence 221 was “less large” in comparison). In short, between-
subjects designs, in which each subject is exposed to one stimulus response, are likely to
have greater unaccounted-for random error, and have been vigorously argued against
(Brunswick, 1956; Birnbaum, 1982).
A major advantage of within-subjects designs, other than the economical
advantage of requiring less subjects since subjects severe in more than one experimental
condition, is that each subject acts as their own control so that consistent error differences
may be isolated amongst different subjects (Weiss, 2006). There is, however, one major
limitation to within-subjects designs: the repetitive nature of a within design could lead
keen subjects to figure out the experimental manipulations which could lead subjects to
give the response they believe the experimenter is ‘looking for’ rather than what the
subject actually believes (Nimh, 1984). In such a case, even significant results would be
spurious and the experimenter would have no way of ascertaining this repetition artifact.
One way to test for this repetition effect is with a Latin Square factorial design. A
Latin Square design is a form of partial counter-balancing where each experimental
condition appears with equal frequency and every possible ordinal position (Weiss,
2006). If subjects complete all the experimental conditions, such a design can be
17
analyzed as a traditional within-subjects factorial design. However, if subjects are
randomly assigned to each ordinal sequence of conditions, each subject’s first response is
conditionally independent and therefore may be analyzed as a between-subjects design.
This approach to testing the repetition effect has yet to appear in the literature on
experimental design. Although not directly related to the substantive hypotheses of this
experiment, the current methodology is used to test the hypothesis that the repetition
effect is apparent within the current context.

18
Chapter 3. Experimental Analysis
Within analysis
Dichotomous decision to commit or release
When the dependent variable is a dichotomous decision to either commit or
release the patient, a 2-way ANOVA detects main effects for framing F(1, 292)=27.78,
p<.001, η
2=
.087, risk level F(2, 292)=1461.67, p<.001, η
2=
.833 and a significant
interaction F(1,292)=14.75, p<.001, η
2=
.048
Table 1. Percentages of Within-subjects Commitment Decisions
Low Medium High
Violent 7.5 (294) 33.1 (293) 91.2 (294)
Non ‐Violent 7.5 (293) 19.8 (293) 78.9 (294)
Cumulative 13.3 (587) 26.5 (586) 85 (588)

% of commit
()=N

Table 1. contains the sample size and within cell percentages of commitment decisions.
The parallelism of the medium and high risk lines in Figure 1. suggests the
significance of the interaction is primarily attributable to low risk level condition. This
level of risk apparently represents a floor effect (the difference between the two groups is
not significant t(584)=0, p=1.0). An orthogonal contrast between the moderate/high risk
groups and framing detects main effects for framing F(1, 292)=38.45, η
2=
.116 and risk
level F(2,292)=599.83, η
2=
.673, but the interaction is not significant F(1,292)<1.

19
Figure 1. Within-subjects proportion of decisions to commit by risk level and
probability frame
0
0.25
0.5
0.75
1
low medium high
Risk probability
Porportion of Decisions to Commit
p(not violent)
p(violent)

Likelihood of commitment or release
Although legal proceedings require a dichotomous decision to be rendered, it is
difficult to gage subjects’ attitudes towards commitment with a dichotomous response.
Moreover, this approach masks subjects’ attitudes with respect to understanding whether
or not they are sensitive to the subtle manipulations in this experiment. Therefore, after
subjects rendered their dichotomous decision, subjects also gave a likelihood rating on a
6-point scale to better convey their attitudes towards commitment. A within subjects
ANOVA detects main effects for framing F(1, 292)=107.03, p<.001, η
2=
.268, risk level
F(1,292)=1299.44,p<.001, η
2=
.817, and a significant interaction F(1, 292)=23.34, p<.001,
η
2=
.022.
20
Figure 2. Within-subjects mean likelihood rating of decision to commit by risk
level and probability frame
0
1
2
3
4
5
6
low medium high
Risk Probability
Estimated Marginal Mean
p(not violent)
p(violent)

Figure 2. indicates that low and high risk have straightforward implications for

commitment decisions, irrespective of the way in which they are framed, but that the

decision to commit or release for the moderate level of risk large depends on the way the

probability is framed. If framed as non-violent, the mean response for moderate risk is

2.56 (S.D. = 1.12; median = 2), and if framed as violent the mean response for moderate

risk is 3.26 (S.D. = 1.29; median = 3). The difference between moderate risk framed as

violent and moderate risk framed as non-violent is significant t(584)=-3.69, p<.001.

21
Between Analysis

Dichotomous decision to commit or release

Table 2. contains the percentages of commitment decisions and sample size for

each condition from the between-subjects analysis with the dichotomous dependent

variable (i.e. commit or release). A 2-way ANOVA detected main effects for framing

F(1, 297)= 32.44, p<.001, η
2=
.098, and risk probability F(2, 297)= 70.52, p<.001, η
2=
.322

as well as an interaction between framing and risk probabilities F(2, 297)= 14.78,

p<.001, η
2=
.091.

Table 2. Percentages of Between-subjects Commitment Decisions

Low Medium High
Violent 2.1 (48) 33.0 (54) 86.7 (45)
Non ‐violent 6.9 (58) 7.1 (42) 37.5 (56)
Cumulative 4.7 (106) 20.05 (96) 62.1 (101)

% of commit
()=N

As Figure 3. indicates, there is a spreading interaction in the difference between
medium and high risk and the way in which the probabilities are framed. There is a slight
crossover interaction in the low risk group where the probability framed as non-violent
yielded a higher number of commitment endorsements than did the probability framed as
violent. A test of orthogonal contrasts indicates the difference between the framing
conditions for low risk is not significant F(1, 106)<1. Orthogonal contrasts of the
differences between the medium/high risk groups and framing still yields a significant
interaction F(1, 197)=3.834, η
2=
.019, albeit a modest one. A test of contrasts indicates the
22
difference between high risk framed as non-violent and moderate risk framed as violent is
not significant t(108)<1.
Figure 3. Between-subjects proportion of decisions to commit by risk level and

probability frame
0
0.25
0.5
0.75
1
low medium high
Risk Probability
Porportion of Decisions to Commit
p(not violent)
p(violent)

Likelihood of commitment or release
Here the dependent variable is a rating on a 6-point scale that reflects the
likelihood of ordering a commitment or release. Results similar to the within subjects
analysis were attained; a 2-way ANOVA detected main effects for framing F(1,
297)=39.75, p<.001, η
2=
.118, and risk level F(2, 297)=133.00, p<.001, η
2=
.473 , as well
as a significant interaction between the two F(2, 297)=7.926, p<.001, η
2=
.051.

23

Figure 4. Between-subjects mean likelihood rating of decision to commit by risk
level and probability frame
0
1
2
3
4
5
6
low medium high
Risk Probability
Mean Likelihood of Commitment
p(not violent)
p(violent)

Figure 4. indicates that a medium risk level framed as violent is likely to result in
a decision to commit (mean = 3.43, S.D. = 1.14), but a medium risk level framed as not-
violent is likely to result in a decision to release (mean = 2, S.D. = .937), a contrast that is
significant (t(94)=-6.72, p<.001)). High risk and low risk had straightforward
implications for the likelihood decisions irrespective of the way in which they were
framed. The mean for high risk framed as violent is 3.42 (S.D. = 1.26) and the mean for
medium risk framed as non-violent is 3.62 (S.D. = 1.35), a difference that is not
significant t(107)=.831, p<.408.
24

Resolution
Taken as a whole, the results suggest both the framing and the risk level are
effectual for commitment decisions. Notwithstanding the floor effect for low risk
probabilities, a significant interaction is still achieved in the between-subjects’
dichotomous decisions. However, to a large extent, the results suggest framing has a
greater effect on subjects’ decisions to commit or release. This is particularly apparent in
Figure 3. and Figure 4. where the difference between medium risk framed as violent and
high risk framed as non-violent is not significantly different. This seems to buttress the
hypothesis that decision makers tend to visualize the error type more than the risk
probability in making decisions, since subjects are willing to commit at the same rate for
unequal probabilities based solely on how the probability is framed. This effect is most
consequential for probabilities corresponding to moderate risk. Both the between-subjects
analysis and the within-subjects analysis of the likelihood of decision reveals that
depending on how it is framed, moderate risk is either sufficient or insufficient to justify
civil commitment. This incoherence is consistent with the hypothesis that probabilities
framed as violent lead subjects to focus on the effects of false negatives, and thus more
likely to commit, whereas probabilities framed as non-violent lead subjects to focus on
the effects of false positives, and thus more likely to release.
This phenomenon can be explained by what Tversky and Koehler (1994) coined
Support Theory, according to which probabilistic judgments tend to be sub and
superadditive based on the ‘unpacking’ or decomposition of uncertain events (Sloman et
25
al. 2004). Although the current study did not ‘unpack’ the event or its probabilities, the
theory applies because its central feature is the effect on the outcome based on the
relative salience of the description of the event; it is of no consequence if this effect is
achieved by the ‘unpacking’ or the framing of the event. Moreover, Support Theory does
not require a probabilistic estimate to be admissible. For example, in the Johnson et al.
(1993) study, subjects indicated the amount of money they would spend on insurance,
based on two different descriptions of the same event. The different description was
consequential for decisions to spend money, not for the perceived probability of the event
occurring. The framing effect is thus a derivative of Support Theory because a different
description of risk level lead to different judgments, although the error committed is even
more fundamental.
An interesting result of the analyses has to do with the methodological contrasts
of the within-subjects and between-subjects designs. One would expect the within design
to yield larger effects since there is less randomness since each subject is their own
control, however, the opposite was found. A visual comparison of Figure 1. and Figure 3.
indicates the effects were more pronounced for the between-subjects design than the
within-subjects design. One interpretation is that the between-subjects results were
inflated due to error variance, and when controlled for (as in the within design), the
interaction disappears and the main effects are more accurate. Another interpretation is
that subjects became aware of the framing manipulation and artificially compensated to
make their decisions coherent. Confidence intervals were included to help obfuscate the
complementary nature of the different risk probabilities, but they may not have gone far
enough; after reading only a few vignettes which explicitly said “will engage in violence”
26
or “will not engage in violence”, subjects could have became keen to the manipulation.
Read in one light, these results offer support for the repetition argument against within-
subjects designs. It is also possible that the inclusion of confidence intervals contributed
to the incoherence because it was more difficult (or less salient) for subjects to calculate
the corresponding complementary probability. Perhaps with a simple point estimate,
rather than a range estimate, subjects could have conceived the complement more
effectively.

27
Chapter 4. Evaluation of Assumptions

Base rate assumptions and putative commitment thresholds
The only empirical research to investigate the legal threshold for commitment
found great variability (Monahan and Silver, 2003). In the study, judges were asked to
indicate “the lowest likelihood of violence to others that you would accept as fulfilling
the ‘dangerousness’ criterion for authorizing short-term civil commitment? (p.3)” The
majority of the judges agreed that 26% is sufficient risk to justify short-term civil
commitment. This result seems surprisingly low since the Supreme Court has noted “civil
commitment constitutes a significant deprivation of liberty” (Addington v. Texas, 1971,
p.1178) which has led commentators to note that civil commitment is reserved for the
most dangerous individuals (Scurich & John, 2009). The magnitude of liberty deprivation
is usually equated in terms of length of incarceration. Although it was undefined in the
Monahan and Silver (2003) study, a 48-hour hold is sometimes referred to as short-term
commitment and is subject to a probable cause standard of proof (Slobogin, 2006). In
California, 48-hour holds are called “emergency detentions” or “psychiatric evaluations”
after which the patient is subject to a “short-term commitment”, which refers to a 14-day
incarceration to a psychiatric facility (Cal. Welfare and Institutions Code, 5150-5250)
and is sometimes subject to a different standard of proof. In California, after the 14-day
commitment, the patient is subject to a 180-day commitment if the person is still
statutorily dangerous. Slobogin (2006) notes: “generally the law requires a lesser showing
of dangerousness as one moves down the hierarchy of interventions (p. 143).” Hence, to
some extent, the low likelihood of violence that the judges purport is unsurprising
because it refers to the bottom of the hierarchy, where mistakes are least costly.
28
The current analysis is concerned with the upper end of the hierarchy: 14-day and
180-day commitments, which are referred to as ‘short-term commitment’ and ‘long-term
commitment’, respectively. In making these definitions explicit, subjects were asked for
the minimal risk probability they consider sufficiently dangerous to authorize
commitment. For short-term commitment, the median risk probability was 60% with
mean 57.25 (S.D. = 21.5), and for long-term commitment the median probability is 80%
with a mean 74.78 (S.D. = 20.52).
The base rate assumptions of subjects are also of interest. Public opinion surveys
indicate that people have ‘exaggerated’ fears of the mentally disordered (Pescosolido et
al. 1999), which has been consequential for legislative policies governing civil
commitment (Monahan, 2006). This implies that people assume a high base rate of
violence amongst the mentally disordered, and that this assumption is incorrect.
Notwithstanding technical issues of time parameters, operationalization and reference
classes, the empirical literature suggests the base rate of violence amongst the mentally
disordered spans a range of 0% with an upper bound of 43% (Rice and Harris, 1995). The
base rate of violence in the Macarthur Study of Mental Disorder and Violence was 18.7%
(Monahan et al., 2001). Subjects in this experiment were asked about their base rate
assumptions. Specifically, they were asked, “if you took a random sample of 100 people
with some type of mental disorder, how many would commit a violent act (any act in
which there is a victim who suffers physical harm)?” It is important to note that the time
parameter is indeterminate, but that violence is operationalized as an act in which there is
a victim who suffers physical harm. Also, the question calls for a frequency, rather than
probability. Given these ascriptions, one would expect the base rate assumptions of
29
subjects to be a moderate-to-upper bound estimate since violence the time parameter is
indefinite and the frequency is claimed to lead to heightened perceptions of risk (Slovic,
Monahan and MacGregor, 2000).
Figure 5. Distribution of Base Rate Assumptions
100 80 60 40 20 0
60
50
40
30
20
10
0
Frequency

Base rate assumption of violence

The mean estimate of violence was 20 (S.D. = 18; median 15). Figure 5. indicates
the base rate assumptions have a strong positive skew. A correlation analysis of base rate
assumptions and putative thresholds reveals the two are significantly and negatively
related. The Pearson correlation of the putative short-term commitment threshold and
30
base rate assumption is -.212 (p<.001). The Pearson correlation for the putative long-term
commitment threshold is -.196 (p<.001).
The putative thresholds elicited from subjects were much greater than those in the
Monahan and Silver (2003) study, and this result is coherent because thresholds in this
study correspond to heightened levels of liberty restriction. The results suggest that on the
whole, subjects are willing to deprive liberty only when risk is greater than 60%. And
that when the sought level of liberty to be deprived increases, so too does the threshold
for commitment. Contrary to the implications of public opinion surveys, subjects did not
express unrealistic base rate assumptions of violence. Although surveys indicate people
have exaggerated fears of mental disorder and violence, it is possible these surveys
conflate the frequency and the magnitude of violence. For example, the public might
believe the magnitude of violence among the mentally disordered is greater than amongst
the normal population, as in the Virginia tech massacre (Cullen, 2007), but the current
results indicate that subjects are sensitive to the anomalous frequency with which this
magnitude of violence occurs. The negative correlation between putative threshold and
base rate assumption tentatively indicates that subjects require more evidence (or a
greater risk) to justify commitment as the occurrence of violence decreases. However,
technically subjects indicated threshold posterior probabilities, which already includes the
base rate assumption; therefore, the results indicate that subjects actually require a
smaller likelihood to commit, since, ceteris paribus, if the posterior does not change but
the base rate increases, the likelihood ratio must decrease. However, it seems plausible
that subjects considered their putative threshold to represent a likelihood ratio threshold
that is insensitive to base rate assumptions.
31

Chapter 5. Discussion

The gravity of civil commitment was noted almost 5 decades ago: “Involuntary
confinement is the most serious deprivation of individual liberty that a society may
impose (Livermore, Malmquist & Meehl, 1968, p.75).” All forms of civil commitment
are premised on forecasts of violence, and the Supreme Court has reaffirmed the
constitutionality of this premise. Moreover, the Supreme Court has asserted the
assessment of violence is an activity in which clinicians are qualified to partake, quite
contrary to the sentiment of the APA. Despite being initially inveighed, subsequent
psychological research on violence risk assessment has been immense, the results of
which have only recently come to fruition. The most conspicuous result is the shift to
actuarial risk assessment.
Actuarial risk assessment relies on empirically established relationships and has
the allure of markedly improving predictive validity. Recent studies have utilized diverse
and comprehensive samples on a scale unfathomable to fledging researchers; it is these
samples on which actuarial risk tools have been developed. At the dawn of their
implementation, concerns regarding the output of actuarial risk assessment spawned an
area of research on the communication of risk, which to this day remains understudied.
Perhaps the most candid realization is that gains in predictive accuracy are superfluous if
decision makers cannot use the results meaningfully (Heilbrun, 1997). It therefore makes
little sense that empirical research has so narrowly focused on improving predictive
validity without regard for the effect it has on decision makers. Empirical research in this
domain has tended to be concerned with decision makers’ preferences for risk assessment
32
(e.g. Heilbrun, O’Neil, Stevens, Shroman & Yi-Wen, 2004). This line of research shows,
not surprisingly, that judges are skeptical of actuarial results and prefer the conventional
testimony of clinicians (Redding, Floyd & Hawk, 2001). This skepticism might come
from a general disdain for probabilities; after all, as one famous law professor has pointed
out: “lawyers are wordsmiths, not number crunchers (Kaye, 1979, p. 34).” But, given the
current direction of research on violence risk and the advances it confers, innumeracy
does little to absolve judges of their duty to evaluate risk. The actuarial approach simply
makes their decisions more transparent and therefore more open to scrutiny.
Heilbrun et al (1999) observe that the form of risk communication has the
potential to ‘mislead’ and that this avenue has yet to be studied. To understand this
observation, ‘mislead’ must be contextualized and evaluated in terms of its relevant
purpose. The purpose of risk communication, and risk assessment more broadly, is to best
inform a legally empowered decision maker in making a decision that entails balancing
the cost of false positives and false negatives (Monahan & Wexler, 1978). To ‘mislead’
would be to provide information that unfairly biases the task of balancing false positives
with false negatives. The results reported herein support the fear that the communication
of actuarial risk probabilities is misleading. Empirical research from broad domains
indicates that human decision makers tend to violate extensionality, and, apparently, the
field of violence risk assessment is no exception. Risk probabilities that are
communicated in frame of the probability of being violent, as is convention, tend to
emphasize false negatives, whereas the probability framed as being non-violent tends to
emphasize false positives, and these frames are consequential for commitment decisions--
particularly when the risk probability is diffuse. Put simply, the communication of risk
33
probabilities is misleading because the frame emphasizes one type of error and this
effects commitment decisions. This is not an argument for the categorization of risk
probabilities; rather it is a call for further research. Further research, for example, might
explore the effects of presenting the risk probability framed both ways, as the probability
of being violent and the probability of being non-violent, to see if this cancels out the
bias towards false positives or false negative. This is very realistic since the counsel for
the respondent could counter a risk probability estimate by explicating the complement of
that risk probability (for e.g. if the client has a 26% probability of violence, his attorney
could respond that means there is a 74% chance of non-violence).
Another deficiency in the empirical literature, and the on which the vitality of the
actuarial approach depends, is whether decision makers are sensitive to differential risk
probabilities. The monotonicity of the reported results suggests decision makers can
appraise different probabilistic levels of risk, and that the differences are meaningful in
decisions for commitment. The probabilities used in the study correspond to risk groups
from the COVR, the superior actuarial tool currently available. A potential caveat is that
the disparity between the probabilities was large with relatively straightforward
implications for commitment decisions. For example, presumably most would consider
the crude labels of low/medium/high risk to validly characterize the level of risk, even
though such a label is itself an evaluative judgment, since there is no official consensus
on what constitutes “low” risk versus “medium” risk, etc. Moreover, such a
characterization is likely to have straightforward implications for commitment
decisions—at least for low and high risk. This implicates two separate but important
issues.
34
The first is a concern that the straightforward nature of the risk probabilities could
make the observed monotonicity an artifact, since decision makers could be aware of the
ordinal relationship between the varying levels of risk. This is a reasonable conjecture,
but there are reasons to believe this is not the case. First, in addition to there being six
different risk estimates (three risk levels framed two different ways), there were
confidence intervals around all six estimates, which means subjects were exposed to
eighteen different probabilities. The exposure to so many different probabilities should
help to obfuscate the exact relation amongst the probabilities. Also, the substantive
results from the between-subjects analysis and within-subjects analysis are consistent.
Decision makers who were exposed to only one risk probability with a confidence
interval (the between subjects analysis) were sensitive to the magnitude of varying risk
levels, thus there is no reason to assume that the results of the within-subjects analysis
reflects an artifact based on an ordinal awareness of risk probabilities.
The second issue has to do with the ongoing debate about the apposite form of
risk communication. Despite the near unanimity that the probabilistic form is most apt,
there are still respected researchers that argue otherwise (Douglass and Reeves, 2009).
These researchers argue that current state of science does not permit valid numeric
estimates because of the sample-specific nature of actuarial probabilities as well as the
nomothetic-idiogrpahic concerns that probabilities derived from the group level might not
apply at the individual level (Douglas and Reeves, 2009, p. 20-29). Instead, these
researchers suggest a holistic approach to evaluation in which the patients are categorized
as high, med or low risk based on clinical judgment. They note the task of categorizing,
“is almost peripheral to the main task at hand…it is a short-hand way to note the person
35
is likely to be violent and in need of services… it is a vehicle through which evaluators
express their judgments about case prioritization (Douglass and Reeves, 2009, p. 22).”
Although the underlying logic for their approach was borne out of common actuarial
criticisms, their approach is not exempt from the same shortcomings. For example, the
nomothetical-idiogrpahic issue is not resolved, as these commentators contend, by
allowing clinicians to modify and heuristically categorize patients. If such categorization
is not based on sufficient data (i.e. a sample), it is merely stereotype and likely to fall
prey to the same biases of traditional clinical judgment. All assessment tools are based on
the group level analysis, and the integration clinical judgment does not transcend this
gap. Not only do the authors concede, “there are unfortunately few studies that have
investigated reliability [of their approach] (p. 29),” the authors fail explain how their
clinically categorized judgments are to be utilized by a legal decision maker. A more
complete critique of this approach can be found in Qunisey et al. 2006.
Aside from the scientific arguments, it appears that judges have a clear preference
for categorical estimates of risk (see Kwartner, Lyons & Boccaccini, 2006). A
consideration of the logical relationship between risk assessment and commitment
decisions reveals the categorical format is objectionable in principle, as previously noted.
Again, there is no scientific consensus on what probability constitutes “high risk” or
“medium risk”, etc., but even if there were consensus, the fundamental problem would
still exist. To the extent the risk categories have clear implications for commitment
decisions, for example low risk is released, high risk is committed, categories usurp the
appropriate role of the clinician because they extend the role from description to
evaluation. The values (false positive and false negative tradeoff) of the clinician do not
36
necessarily encompass the constitutional values on which civil commitment is legally
justified, especially when clinicians have monetary and professional interests in seeking
commitment. Furthermore, clinicians can mitigate tort liability by proffering an
assessment rather than an evaluation, since an evaluation or decision would be the
responsibility of someone else. In sum, clinicians should refrain from proffering
categorical estimates of risk out.
There is a real tension in the use of categorical risk estimates between the
practical and the prescriptive. However, the current results indicate this entire debate is
avoidable because decision makers can use risk probabilities meaningfully. Although
there are issues with the way in which the probabilities are framed, decision makers are
sensitive to differences in the magnitude of risk. By communicating risk as a probability,
clinicians are confined to their proper role of description and decision makers are
confined to their proper role of evaluation. In such a paradigm, legal decision makers
cannot simply defer to the evaluation of the clinician.
Further research should focus on intermediary risk probabilities, as these are the
most difficult to evaluate. In addition, there ought to be some discussion on the
appropriateness of clinicians reporting posterior probabilities as risk probabilities. It has
been argued that expert witnesses should only convey likelihood ratios, and the judge or
some other source should supply the prior odds (Wagenaar, 1988). This approach has
been suggested within the context of civil commitment as well because base rates
inherently vary across location and context (Scurich & John, 2009). If this approach were
to come to fruition, further research should examine the efficacy with which judges
37
combine these components, and possible the effects of pedagogy on legal decision
makers.

38
Bibliography

Anderson, N.A. (1981) Foundations of information integration theory. Academic Press:
New York

Baron, J. (2000) Thinking and deciding (3
rd
ed.). Cambridge University Press: New York,
NY

Birnbaum, M. H. (1982). Controversies in psychological measurement. In B. Wegener
(Ed.), Social attitudes and psychophysical measurement (p. 401-485). Hillsdale,
NJ: Erlbaum.

Birnbaum, M.H. (1999) How to show that 9>221: Collect judgments in a between-
subjects design. Journal of Methods, 4, 243-249

Brunswik, E. (1956). Perception and the representative design of psychological
experiments (2nd ed.). Berkeley: University of California Press.

Carlsmith, K., Monahan, J. and Evans, A. (2007) The function of punishment in the
“civil” commitment of sexually violent predators. Behavioral Sciences and the
Law, 41, 224-239

Clemen, R. (1991) Making hard decisions. An introduction to decision analysis. PWS-
Kent: Boston

Cohen, J. & Cohen, J. (1977) Statistical power analysis for the behavioral sciences.
Academic press: New York

Dawes, R.M., Faust, D., & Meehl, P.E. (1989) Clinical vs. actuarial judgment. Science
243, 1668-1674

Douglass, K.S. & Reeves, K. (2009) HCR-20 Violence assessment scheme: rationale,
application and empirical overview

Fisher, R.A. (1925) The design of experiments. Hafner: New York

Gigerenzer, G. & Hoffarge, U. (1995) How to improve Bayesian reasoning without
instruction: frequency format. Psychological Review, 102, 678-704

Grove, W., & Meehl, P. (1996) Comparative efficiency of informal (subjective,
impressionistic) and formal (mechanical, algorithmic) prediction procedures: The
clinical statistical controversy. Psychology, Public Policy and Law (2)2, 293-323

Grove. W., Zald, D.H., Lebow, B.E., Snitz,B., & Nelson, C. (2000) Clinical versus
mechanical prediction: meta-analysis. Psychological Assessment 12, 19-30
39

Harris, G., Rice M. Quinsey, V. (1997) A multi-site comparison of actuarial risk
instruments for sex offenders. Psychological Assessment, 15, 413-425

Heilbrun, K. (1997) Prediction versus management models relevant to risk
assessment: the importance of legal decision making context. Law and Human
Behavior, 21, 347-359

Heilbrun, K., Dvoskin, J., Hart, S., & McNiel, D. (1999). Violence risk communication:
implications for research, policy, and practice. Health, Risk & Society, 1, 91-106.

Heilbrun, K., O’Neil, M., Stevens, T., Strohman, L., Bowman, Q., & Lo, W. (2004)
Assessing normative approaches to communicating violence risk: a national
survey of psychologists. Behavioral Sciences and the Law, 22,187-196

Heilbrun, K., O’Neil, M., Stevens, T., Strohamn, L.K., Bowman, Q., & Yi-Wen, Lo
(2004b). Assessing normative approaches to communicating violence risk: a
national survey of psychologists. Behavioral Sciences and the Law, 22, 187-196

Johnson, E.J., Hersery, J., Meszaros, J. & Kunreuther, H. (1993) Framing, probability
distortions, and insurance decisions. Journal of Risk and Uncertainty, 7, 35-51

Kaye, D.H. (1978) The laws of probability and the laws of the land. University of
Chicago Law Review, 47, 34-56

Kwartner, P., Lyons, P.M. & Boccaccini, M.T. (2006). Judges’ risk communication
preferences in risk for future violence cases. International Journal of Forensic
Mental Health, 5, 185-194

Livermore, J., Malmquist, C.P., Meehl, P. E. (1968) On the justifications for civil
commitment. University of Pennsylvania Law Review, 117, 75-96

Meehl, P. E. (1954) Clinical versus statistical prediction: A theoretical analysis and
review of the evidence. Minneapolis: University of Minnesota Press.

Melton, G., Petrila, J., Poythress, N., & Slobogin, C. (1987) Psychological evaluations
for the courts. New York: Guilford.

Monahan, J., & Wexler, D. (1978) A definite maybe: proof and probability in civil
commitment. Law and Human Behavior 2(1), 37-42

Monahan, J. (1981) The clinical prediction of violent behavior. DC: Government
Printing Office

40

Monahan, J., & Steadman, H.J. (1994) Violence and mental disorder: Developments in
risk assessment. Chicago: The University of Chicago Press

Monahan, J., & Steadman, H.J. (1996) Violent storms and violent people: how
Meteorology can inform risk communication in mental health law. American
Psychologist, 51, 931-938

Monahan, J., Steadman, H., Silver, E., Appelbaum, P., Robbins, P., Mulvey, E., Roth, L.,
Grisso, T., & Banks, S. (2001) Rethinking risk assessment. New York: Oxford
University Press.

Monahan, J., Heilbrun, K., Silver, E., Nabors, E., Bone, J. & Slovic, P. (2002)
Communicating violence risk: frequency formats, vivid outcomes, and forensic
settings. International Journal of Forensic Mental Health 1(2),121-126

Monahan, J., & Silver, E. (2003) Judicial decisions thresholds for violence risk
management. International Journal of Forensic Mental Health 2, 1-6

Monahan, J. (2005) Clinical and actuarial predictions of violence: scientific status. In
Faigman, D., Kaye, D., Saks, M. & Sanders, J. (eds.) Modern scientific evidence
(90-109) Massachusetts: Thompson West

Monahan, J. (2006) A jurisprudence of risk assessment: forecasting harm among
prisoners, predators, and patients. Virginia Law Review 92(3)

Monahan, J. (2008) Structured risk assessment of violence. In Simon, R. & Tardiff, K.
(Eds.) Textbook of violence assessment and management. (17-33) Washington
D.C.:American Psychiatric Association

Mulford, M. & Dawes, R.M. (1999) Subadditivity in memory for personal events.
Psychological Science, 10, 47-51

Nihm, Sue-Doe (1984). Self-reports on mental processes: A response to Birnbaum and
Stegner. Bulletin of the Psychonomic Society, 22, 426—427.

Nisbett, R.E., & Ross, L. (1980) Human inference: Strategies and shortcomings of social
judgment. New Jersey: Prentice-Hall

Pescosolidio, B., Monahan, J., Link, B. Stueve, A., and Kikuzawa, S. (1999) The public’s
view of the competence, dangerousness, and need for legal coercion of person
with mental health problems. American Journal of Public Health, 9, 1339-1345

41

Qunisey, V.L., Harris, T.G., Rice, G.T. & Comier, C.A. (2006). Violent offenders:
Appraising and managing risk. American psychological Association:
Washington, D.C.

Redding, R.E., Floyd, M.Y. & Hawk, G.L. (2001) What judges and lawyers think about
the testimony of mental health experts: A survey of the courts and bar. Behavioral
Sciences and the Law, 19, 583-594

Rottenstreich, Y. & Tversky, A. 91997) Unpaking, repacking and anchoring: advances in
support theory. Psychological Review, 104, 406-415

Schopp, R., & Quattrocchi, M.R. (1995). Predicting the present: expert testimony and
civil commitment. Behavioral Sciences & the Law, 13, 159-181

Schopp, R. (1996) Communicating risk assessments: accuracy, efficacy and
responsibility. American Psychologist, 51, 939-944

Scurich, N. & John, R. (2009). The normative threshold for psychiatric civil commitment.
Forthcoming.

Slobogin, C. (2006) Minding justice: Laws that deprive people with mental disorder of
life and liberty. Massachusetts: Harvard University Press

Sloman, S., Rottenstreich, Y., Wisniewski., E., Hadjichristidis, C., Fox, C. (2004) Typical
versus atypical unpacking and superadditive probability judgment. Journal of
Experimental Psychology, 30, 573-582

Slovic, P., & Monahan, J. (1995) Probability, danger, coercion. A study of risk
perception and decision making in mental health law. Law and Human Behavior,
19, 49-65

Slovic, P., Monahan, J., & MacGregor, D. (2000) Violence risk assessment and risk
communication: The effects of using actual cases, providing instruction, and
employing probability versus frequency formats. Law and Human Behavior, 24,
271-296

Szasz, T. (2003) Liberation by oppression. A comparative study of slavery and
psychiatry. Transaction: New York

Tversky, A. & Koehler, D. (1994) Support theory: a nonextensional representation of
subjective probability. Psychological Review, 101, 547-567

Underwood, B. (1979) Law and the crystal ball: Predicting behavior with statistical
inference and individualized judgment. Yale Law Review, 88, 1421–22
42

Vose, D. (2000) Risk analysis: A quantitative guide. John Wiley and Sons: New York

Wagenaar, W.A. (1988) The proper seat: A Bayesian discussion of the position of expert
witnesses. Law and Human Behavior, 12, 499-510.

Weiss, D.J. (2006) Analysis of variance and functional measurement. Oxford University
Press: New York

Webster, C., Douglas, K. & Eaves, D. (1997) HCR-20: Assessing risk for violence.
Vancouver, BC, Canada

Winick, B. (2001) The civil commitment hearing: applying the law therapeutically. In
Frost, L. & Bonnie, J (Eds.) The evolution of mental health law (291-308).
Washington D.C.: American Psychological Association

43
Appendix
California has a legal statue that allows people to be admitted to a mental hospital
against their will if they meet two conditions: 1. they have a mental disorder and 2. are a
danger to self or others. This statue is like quarantine in that it is designed to prevent
violence before it happens. Determining who is dangerous is a difficult task, and research
suggests clinicians cannot accurately make such determinations. Recently, a statistical
approach has been developed that can predict the likelihood that a given patient will be
violent, and with much better accuracy; the statistical tool is called the Classification of
Violence Risk (COVR). Much in the same way insurance companies use a
statistical formula to calculate driving risk, The COVR uses a host of clinical factors to
derive a risk estimate of violent behavior.
Once this risk estimate is known, a judge must decide whether the risk is
sufficiently dangerous to justify committing a person to a mental hospital against his or
her will. The issue is complicated because the person has not broken any laws, which is
usually the necessary prerequisite to having liberty (freedom) taken away. Therefore, the
judge must weigh the rights of the individual against possibly preventing violence in the
community.
You are asked to assume the role of a judge. Your task is to determine if a given
patient is sufficiently dangerous to justify an involuntarily commitment to a mental
hospital. You should assume the “mental disorder” criterion is satisfied. Among other
things, your decision should consider: a) the individual’s right to liberty (freedom) b) the
protection of society from harm c) the cost of incarceration to society d) the
44
stigmatization to the individual and his/her family. You are going to read 6 brief case
summaries that were prepared by a clinician whose professional opinion you trust; these
are real cases and involve actual people being considered for commitment. After reading
each report you will be asked to make a decision to either I.) order a 14-day commitment
or II.) release the patient, with or without further conditions. Please take your time in
thinking about each case and imagine your decisions will have a real live effect on the
patients in question.
Case Example
Mr. Smith is a 27-year-old male salesman who has been hospitalized for the
eighth time with a diagnosis of bipolar disorder. After 72 hours in the hospital, he is
being considered for a certification of commitment. Since an aggressive act toward Mr.
Smith’s wife while manic and intoxicated had precipitated his hospitalization, the
clinician responsible for treatment requests that the COVR be administered. The next
day, a COVR report concludes: “The likelihood that Mr. Smith will not commit a violent
act toward another person is estimated to be between 14 and 35 percent, with a best
estimate of 24 percent.” The report also lists the risk factors that were used to produce
this estimate.
The clinician, after reviewing the COVR report and all the information in Mr. Smith’s
hospital chart, interviews Mr. Smith. The interview fails to uncover any unusual
protective factors that would call into question the estimate of violence risk that COVR
had produced. Since his manic state has not fully resolved, the clinician decides to
continue a course of medication and anger management therapy.
45
46
Do you consider this patient to be sufficiently dangerous to order a 14-day
commitment to a mental hospital?

Abstract (if available)

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

The dynamics of reasonable doubt

PDF

The influence of risk on decision-making during walking

PDF

Effects of stress on teacher decision making

PDF

Discussion during treatment decision-making predicts emotional adjustment in prostate cancer patients

PDF

Metacognitive experiences in judgments of truth and risk

PDF

A two-level analysis of foreign policy decision making: an empirical investigation of the case of China-Taiwan

PDF

Investigating the effects of personality factors on goal shielding

PDF

Deconstructing the psychological components of emotional decision making and their relation to the suicide continuum

PDF

Direct and indirect predictors of traumatic stress and distress in orphaned survivors of the 1994 Rwandan Tutsi genocide

PDF

Emergent literacy skills and home literacy environment of Latino children who live in poverty

PDF

Taking the temperature of the Columbia Card Task

PDF

The effects of familism and cultural justification on the mental and physical health of family caregivers

PDF

Antecedents of marijuana initiation

PDF

When it's good to feel bad: how responses to virtual environments predict real-life sexual risk-reduction

PDF

The effect of present moment awareness and value intervention of ACT on impulsive decision-making and impulsive disinhibition

PDF

Pushmi-pullyu and little c: a search for the structure of personal creativity in a general population

PDF

A discussion of inter-rater reliability and content validity of Westphal's decision making inventory (1967), describing developmental processes of adults with developmental delays

PDF

Essays on the empirics of risk and time preferences in Indonesia

PDF

Marital conflict and parenting: moderating effects of ethnicity and parent sex

PDF

A comparison of standard and alternative measurement models for dealing with skewed data with applications to longitudinal data on the child psychopathy scale

Asset Metadata

Creator Scurich, Nicholas Ivan (author)

Core Title The effects of framing and actuarial risk probabilties on involuntary civil commitment decisions

School College of Letters, Arts and Sciences

Degree Master of Arts

Degree Program Psychology

Publication Date 10/27/2009

Defense Date 09/30/2009

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag actuarial risk assessment,civil commitment,dangerousness,juridical decision-making,OAI-PMH Harvest,violence risk communication

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor John, Richard S. (committee chair), Lyon, Thomas D. (committee member), McArdle, John J. (committee member)

Creator Email nickscurich@hotmail.com,scurich@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-m2694

Unique identifier UC1268318

Identifier etd-Scurich-3357 (filename),usctheses-m40 (legacy collection record id),usctheses-c127-266269 (legacy record id),usctheses-m2694 (legacy record id)

Legacy Identifier etd-Scurich-3357.pdf

Dmrecord 266269

Document Type Thesis

Rights Scurich, Nicholas Ivan

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Repository Name Libraries, University of Southern California

Repository Location Los Angeles, California

Repository Email cisadmin@lib.usc.edu