low. Despite these developments, other legal decisions
upheld the constitutionality of clinical predictions of
violence. Furthermore, the 1976 groundbreaking case of
Tarasoff v. Regents of the University of Californiawent
so far as to impose a positive legal duty on psychiatrists
and psychologists to forecast the potential violence of
patients under therapy in some circumstances.
In the early 1980s, Professor John Monahan summa-
rized what he called the “first generation” of empirical
studies on the prediction of violence. He concluded that
clinical predictions were not very accurate, in particu-
lar leading to an unacceptably high false-positive error
rate. He also called for a “second generation” of
research that would focus on identifying meaningful
risk factors for violence and using empirically based
procedures for making risk assessments. Since that
time, researchers have indeed identified numerous
meaningful violence risk factors, such as substance use
problems, psychopathic personality features, anger,
impulsivity, antisocial peers, antisocial attitudes, previ-
ous violence, young age at first violent act, stress, treat-
ment nonadherence, lack of social support, and some
features of mental illness.
Although there has been no well-recognized decla-
ration of a “third generation” of risk assessment
research, efforts since the early to mid-1990s have cap-
italized on what was learned from risk factor research
that has led to the construction and evaluation of risk
assessment measures that compile and integrate numer-
ous empirically validated risk factors. A further move-
ment has included increased emphasis on the reduction
of risk or prevention of violence, as opposed to solely
estimating the likelihood of future violence. For this
reason, a single definition of violence risk assessment is
elusive. The two primary contemporary approaches to
risk assessment, discussed below, adopt somewhat dif-
fering conceptualizations of the task.Contemporary Approaches to
Violence Risk Assessment
There are two primary approaches to violence risk
assessment—structured and unstructured. The structured
risk assessment typified first-generation research on
risk assessment and remains commonly used today.
However, unstructured risk assessment, sometimes
called clinical prediction, is based primarily on clini-
cians’ discretion and lacks rules that guide the risk factor
selection or integration process. As such, it is vulnerable
to decisional biases and widely varying quality acrossclinicians. For these reasons, although research supports
that it can achieve statistically significant predictive lev-
els, this differs across clinicians. Furthermore, most
research indicates that it has lower reliability and predic-
tive validity than more structured approaches to risk
assessment. Therefore, a purely unstructured, discretion-
based approach to risk assessment cannot form the basis
of defensible risk assessment.
To increase the reliability and validity of risk
assessments, researchers focused on developing and
evaluating structured approaches:actuarial decision
making andstructured professional judgment (SPJ).AAccttuuaarriiaall DDeecciissiioonn MMaakkiinngg
The actuarial approach has a long history in psychol-
ogy in terms of prediction. It is defined by the applica-
tion of algorithms (equations, score cutoffs, decision
rules) to the combination of risk factors to reach a pre-
dictive decision, thus improving the consistency and
accuracy of such decisions. It also tends to incorporate
risk factors that have been selected through empirical
means (i.e., those items that add independently and
incrementally to the prediction of the outcome).
Therefore, at least within development samples, the risk
factors used to make decisions have empirical support.
Meta-analytic research suggests that actuarial prediction
is more accurate than unstructured clinical prediction in
approximately 50% of research studies, by about a 10%
increase in hit rate.
Problems have been noted with actuarial approaches.
Perhaps most important, actuarial measures that rely on
statistical selection of risk factors and decision algo-
rithms are subject to predictive degradation when used
in new samples, and hence their generalizability is
potentially tenuous. Therefore, cross-validation is a vital
component of actuarial measure development, and eval-
uation and must be performed prior to the use of such
measures. Furthermore, there is a tendency for actuarial
methods to emphasize time-invariant risk factors that
are less relevant to violence risk management and reduc-
tion aims than time-varying, or dynamic, risk factors.
Because there has been a strong conceptual shift in the
violence risk assessment field from a purely predictive
model to a risk management or harm-reduction model,
the predominantly predictive focus of some actuarial
methods has been criticized. Although identifying risk
level can indicate the intensityof necessary interven-
tion,the failure to include risk factors that can serve
as treatment targets reduces the relevance of actuarial850 ———Violence Risk AssessmentV-Cutler (Encyc)-45463.qxd 11/18/2007 12:44 PM Page 850