The prediction of violence occupies a prominent and controversial place in public mental health practice. Productive debate about the validity of violence predictions has been hampered by the use of methods for quantifying accuracy that do not control for base rates or biases in favor of certain outcomes. This article describes these problems and shows how receiver-operating characteristic analysis can be used to solve them. The article also reanalyzes 58 data sets from 44 published studies of violence prediction. Taken together, these data strongly suggest that mental health professionals' violence predictions are substantially more accurate than chance. Short-term (1-7 day) clinical predictions seem no more accurate than long-term (> 1 year) predictions. Past behavior alone appears to be a better long-term predictor of future behavior than clinical judgments and may also be a better indicator than cross-validated actuarial techniques.

Receiver operating characteristic (ROC) analysis traditionally has dealt with dichotomous diagnostic tasks (e.g., determining whether a disorder is present or absent). Often, however, medical problems involve distinguishing among more than two diagnostic alternatives. This article extends ROC concepts to diagnostic enterprises with three possible outcomes. For a trichotomous decision task, one can plot a ROC surface on three-dimensional coordinates; the volume under the ROC surface (VUS) equals the probability that test values will allow a decision maker to correctly sort a trio of items containing a randomly-selected member from each of three populations. Thus, the VUS summarizes global diagnostic accuracy for trichotomous tests, just as the area under a ROC curve does for a two-alternative diagnostic task. Information gain at points on the surface can be calculated just as is done for two-dimensional ROC curves, and investigators can thus compare three-way ROCs by comparing maximum information gain on each ROC surface.

Several studies over the past decade have shown that simple rating scales can accurately rank sex offenders' long-term risk of recidivism. But when using these scales as prediction tools, evaluators often wish to translate categories of risk into probabilities of recidivism. D. M. Doren (2004) has recently suggested that evaluators may use the recidivism percentages published in original studies of the RRASOR and STATIC-99 without regard to differences in populations or base rates. This article explains why Doren's computations should lead to a different conclusion, and describes how simply comparing percentages across studies can mislead researchers and clinicians. Instead, investigators should isolate and examine the detection properties of risk assessment instruments alone, independent of the population- or setting-specific base rate. This article explains this process, using an imaginary study to illustrate how base rates and the properties of risk assessment instruments yield estimated probabilities of recidivism. The article also shows why Doren's results imply that the percentages of recidivism associated with scores on the RRASOR and STATIC-99 scores may vary across study populations. The article offers recommendations for researchers who design and evaluate actuarial methods of assessing risk and for clinicians who interpret results from risk assessment instruments.

In their work on the MacArthur Treatment Competence Study, Paul Appelbaum, Thomas Grisso, and their colleagues warn that their "experimental measures" of decisional capacity "should not be interpreted as though they provide determinations of legal incompetence to consent to treatment." The authors of this article do not believe that Appelbaum et al.'s admonition is strong enough, and they identify and analyze the serious generic, inherent problems connected with any attempt to construct a universally acceptable version of a capacimeter. They suggest that the continuing search for the elusive "Holy Grail" test of capacity proceed with great caution (if at all), and they conclude by urging that investigators and scholars devote their energies toward the development and dissemination of appropriate clinical practice parameters.In the United States 1 at the end of the 20th century, medical decision making is predicated on the legal doctrine, 2 as well as the ethical principle, 3 of informed consent. The informed consent doctrine, in turn, is based on the assumption of a decision-making process that is voluntary, that is informed by an adequate disclosure of information to the patient or surrogate decision maker, and that involves an ultimate decision maker who has adequate cognitive and emotional capacity 4 to