 |
 |
 |
Assessing Dangerousness A young man is angry at his wife, frustrated at her erratic behavior and her limiting his involvement with their infant daughter. Out of spite, he takes the child to an adoption agency while she is at work. When she returns, he refuses to tell her where the baby is. He threatens her with a loaded gun. The wife, who speaks limited English, tells a neighbor that her husband said something about the child being the devil. They call the police. This sounds like a dangerous guy. He certainly seems unstable, has a bad temper, and is possibly mentally ill. But are these factors related to dangerousness? What is the probability that he will harm someone in the future? Defense attorneys may want to present favorable information about their clients' future dangerousness at a number of points in the adjudication process (see Table 1). Only twenty years ago, psychology and psychiatry publicly declared that future acts of violence could not be predicted. Early findings indicated that when mental health professionals predicted that someone was dangerous, they were wrong two-thirds of the time. However, more recent studies have lead to better methods and more optimistic appraisals. One problem with early studies of violence prediction was that many such acts went undetected. Studies in the 1970's typically relied on re-arrest or re-hospitalization records, which severely underestimates the rate of violent acts. More recent studies use not only official records, but also reports from the subject and others that know him or her in the community. When measured in this way, the incidence of violence is four to five times greater than in previous studies. It seems that maybe mental health professionals weren't over-predicting dangerousness after all.
However, it appears they were focusing on the wrong risk factors. Clinicians, research shows, judge dangerousness using the same cues as lay persons. For example, they are heavily influenced by the index offense, by mental illness, by low IQ, and are impressed by professions of remorse. Many of these are predictive of future risk, but in the opposite direction you might initially think. It is no longer sufficient for risk assessments to be based on common-sense notions of what makes someone dangerous. There is a substantial research basis for these assessments and it is indispensable for a competent appraisal. Like the defendant above, some clients appear more dangerous from commonsense (but incorrect) notions than when assessed on empirically valid data. Conversely, some sex offenders are now known to be an intractable. This wasn't always common knowledge. Pedophilia was once considered a neurotic (conflicted) act fully amenable to psychotherapy, like just about everything else. Table 2 lists some empirically valid risk factors, as categorized by one popular risk assessment instrument, the HCR-20 (discussed below). Lastly, the task of risk assessment has been reconceptualized. The risk assessor is assessing risk, or probability, that a person will act in violent manner in the future. This is different that predicting a person will be violent, as when some professionals asserted that they were "one hundred percent sure" a person would kill again. The new language recognizes that some low risk people may face unusual, provocative situations, and some high-risk persons may avoid them. Although most risk assessment items focus on the individual and his habits, environment and situation also matter. For example, psychiatric patients that return to a crime-ridden neighborhood have a much higher incidence of recidivism, and this factor may account for the higher rates reported for African-Americans (Silver, Mulvey, & Monahan, 1999). Table 2 - Predictors of Violence (items listed are from the HCR-20, discussed below)
Assessment Collecting and reviewing the raw data for a risk assessment may easily entail four to six hours of professional time. Once data is collected, it must be organized. There are currently three widely researched and well-validated approaches to appraising future dangerousness. Each is based on very different assumptions. Psychopathy CheckList - Psychopathy is the hard kernel of antisocial personality disorder. While about 75% of persons in prison are antisocial, only about 25% are psychopathic. Pure psychopaths are self-centered, manipulative, impulsive and ruthless, as well as versatile criminals. The Psychopathy Check List specifies twenty such Characteristics (see Table 3). After all the data for an assessment is collected, the examiner rates the subject on each factor, on a scale of 0 (not present) to 2 (clearly present). A total score of 40 is possible. Different researchers have utilized scores of 25 or 30 as the cutoff for Psychopathy. High scorers are associated with four times the risk of future violence than low scorers. High scores also predict supervision failure and poor response to psychological treatments. In fact, there is some evidence that psychopathic offenders are made more likely re-offend by traditional treatments, including milieu therapy and sex offender treatment (Quinsey, Harris, Rice, & Cormier, 1998). This may be because psychopaths are very interested in reading people, and group therapy gives them a great deal of access to people speaking unguardedly about their true feelings and how they make decisions. It also provides a model for how to impress therapists who are in a position to evaluate their progress. The Psychopathy CheckList represents a personality approach to future dangerousness. Most of the items are either traits or descriptions of lifestyle. There is some evidence that psychopathy is a taxon--a discrete personality type. This is very rare among personality traits and other psychological vairables, which are assumed to be continuous (like height and weight). The PCL is not only one of the best validated measures, but is incorporated as a scored item in the other three dangerousness appraisal guides discussed next. Even among psychiatric patients, higher PCL-R scores are robustly correlated with higher level of violence. Table 3 - Psychopathy Check List Items
The HCR-20 is a checklist of twenty factors that empirically related to increased risk of violence. Unlike the PCL-R, its sole purpose is to assess of the risk of future violent behavior and plan intervention. The HCR-20 is divided into three sections. The first pertains to "historical" items-features of the person's life history that will not change over time, such as age at first arrest. The second section contains five items that address "clinical" issues, such as lack of psychological insight, symptoms of mental illness, and poor response to treatment. The last section includes factors that may raise the risk of violence, such as the presence of stress and a history of failure in supervision. Each item is rated on a scale from 0-2, with some guidance from the scoring manual. Although items are totaled in each domain, the total score is not referenced to any norms. The clinician is given the task of integrating the data and reviewing how the defendant's life circumstances, and particular history of violence, will translate into future behavior. A summary judgment of "Low," "Moderate," or "High" risk is the suggested way of conveying the overall results of this analysis. However, clinicians are encouraged identify situations and other factors that are relevant to the individual being assessed and which may raise or lower risk. Unlike the other two approaches, the HCR-20 is focused on the clinical management of dangerousness as well as its assessment. Like the HCR-20, The Violence Recidivism Assessment Guide (Quinsey, Harris, Rice, & Cormier, 1998) is a dedicated risk assessment device. However, its philosophical approach is diametrically opposed to that of the HCR-20. The VRAG is based on actuarial prediction, much as auto insurance rates are. In the HCR-20, the clinician remains the final judge of dangerousness. Although numbers are given to code the data, the clinicians may choose not to add them up. In the VRAG, the numbers, once assigned to items, completely determine the estimate of risk. The clinician has no further role other than to report the results with appropriate caveats. The VRAG authors cite numerous studies that have compared the ability of various people, including psychologists, to predict behavior compared to mathematical algorithms. Most times, the algorithm wins (see Grove & Meehl, 1996). Because the scoring rules and judgment process are fixed (do not vary from one psychologist to another), precise risk levels can be associated with a given score. VRAG scores are divided into nine categories, or "bins." The probability of rearrest for a violence offense is 0% for subjects in the lowest bin; for those in the highest bin, the probability is100%. Unfortunately, few patients fall in these most extreme groups, and most will have risk levels between 10% and 80%. When certainty is low, the role of values becomes prominent. Assume someone has a risk of re-arrest for a violent offence of 25% over the next seven years. Although a substantial risk, does this warrant confining four people knowing, on the average, only one will recidivate? These are value judgments, not scientific ones. Actuarial prediction is great in principle, but it has an Achilles heal: The subject whose behavior is to be predicted must be comparable to the subjects used to create the algorithm.1 This is a potential problem in the VRAG, which was developed Canada in a forensic hospital using mostly mentally disordered offenders. Although the sample was large (nearly seven hundred), it was mostly Caucasian and male. The authors argue that their sample is comparable to many correctional settings and that the VRAG's validity generalizes well to subsamples of their study, including sex offenders and psychopaths. The VRAG has also been validated in more diverse populations in the US. The VRAG was created by examining the empirical relation of predictor variables to violence outcome. Because the VRAG variables are considered in relation to each other as well as to the criterion in creating the algorithm, there are some surprising scoring patterns. Although mental illness is a risk factor for future dangerousness, and is listed as such on the HCR-20, a diagnosis of schizophrenia actually reduces the assessment of dangerousness on the VRAG. Why? Because in the population the VRAG authors studied, were two types of people: Crazy people and criminals. When the recidivism rate of the two are compared, criminals are more likely to reoffend than the mentally ill. The VRAG is the most controversial of the three methods. John Monahan, the dean of violence prediction research, wrote: "… for use with male patients with histories of serious violence, the [VRAG] is so far superior to anything else previously available that not to seriously consider its use … would be a difficult choice to justify." On the other hand, Steven Hart, one of the authors of the contributors to the PCL-R and authors of the HCR-20, as well as current president of the Psychology-Law Division of the American Psychological Association, rails against the VRAG and actuarial prediction. At least some of the concerns of the critics have been borne out in recent findings that the VRAG lacks any validity for women. The MacArthur Foundation has done groundbreaking research in the appraisal of future dangerousness. In their most recent project (Monahan, Steadman, Silver, Appelbaum, Robbins, Mulvey, Roth, Grisso, & Banks, 2001), they followed up nearly one thousand psychiatric patients following release into the community. The authors investigated 134 possible predictors and collected data from official records, the patients themselves, and community informants. The PCL-R was the best single predictor, and rates of violent behavior were much higher than observed in most previous studies. Substance abuse was a major risk factor. Traditional approaches to combing predictors resulted in respectable level of prediction. When an innovative multiple classification tree approach was used, however, the authors achieved 50% higher accuracy than typically observed in risk assessment. The approach will likely require software and should be released within the next year. The Texas Department of Criminal Justice Parole Division now uses its own risk assessment instrument when evaluating candidates for parole. Developed by Security Response Technologies, Inc., it consists of a number of indictors of recidivism that are generally consistent with the research literature. However, the only material to document the instrument's validity is an administrative report to the Texas Department of Corrections (McGinnis & Austin, 2001), and this does not include the data necessary to evaluate SRT's claims. Further, the report warns of poor cross-rater reliability on many items and the total score, and proposes a conservative scoring system consisting of 2, 3 and 4 point ratings. This recommendation was apparently ignored by TDCJ, as risk items are now rated on nine and twelve point scales. This probably exacerbates reliability problems and increases the role of subjective judgment. Lastly, SRT found, as did the VRAG authors, that the severity of index offense is inversely related to future dangerousness. Yet, this factor is not included in the SRT risk assessment instrument. In fact, the decision grid for parole gives equal weight to the severity of the offense and the risk score, such that more serious offenses are very heavily weighted against parole even though associated with lower risk.
Table 4 - Parole Guidelines Score
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||