Can a Computer Predict Crime?

1-12-16 National:

Can machines predict where crimes occur before human beings do? Will the technology be used in unbiased ways? How will these devices affect privacy?

As concerns grow over what are called predictive-policing methods and their effect on Americans, one thing is clear: these new crime-fighting tools are not only being implemented around the country; they are increasing in sophistication and scope.

Technology such as PredPol, which runs crime reports through an algorithm to discover locations with a high probability of crime, have already been adopted by police departments in Miami, Seattle, Los Angeles, and Santa Cruz.

The Chicago Police Department in 2013 created a "heat list" which logged 400 people through an algorithm thought to expose future criminals. Its effectiveness and accuracy remain unclear.

In New Orleans, a company called Palantir applied technology to identify approximately one percent of the city population who will be likely to inflict or be victims of gun violence.

The use of such crime-fighting tools has sparked debate among academics, legal experts and privacy advocates.

Law professor Andrew Ferguson of the University of the District of Columbia told a symposium at the Association of American Law Schools annual convention in New York on Friday (January 8) that predictive policing devices can "distort the law" regarding the Fourth Amendment prohibition of illegal search and seizures.

”This [technology] is less about predictions and more about risk,” he said. “It's more about identifying vulnerabilities in a community as opposed to being able to actually predict a particular crime."

Recognizing Offender Needs When Assessing Risk

March 2015:

Offender risk assessments need to include appraisals of destabilizing influences and factors likely to cause criminal behavior — known as criminogenic needs — according to a new study in the journal Criminology and Public Policy.

Researchers examined the offender profiles of 17,252 community-supervised people from North Carolina’s probation system. They found that there were four common classes of “dynamic needs,” which combine both risk and destabilizing influences, such as mental health and substance abuse problems, lack of housing and educational deficits:

• low need with few destabilizers (36 percent)
• moderate need with high destabilizers (28 percent)
• high need with moderate destabilizers (23 percent)
• high need with high destabilizers (13 percent)

The study notes that many jurisdictions limit the eligibility of low-risk offenders for treatment services, but many of the offenders classified as low-risk in the study “were found to have complex dynamic needs.”

“The current study findings indicate that most offenders have service provision needs in multiple life domains. The high prevalence of destabilizers (e.g., mental health and housing) across the classes heightens the importance of integrating case-management services within treatment programming and stresses the necessity for the development of interventions that are sensitive to the complexity of offender treatment needs,” researchers wrote. ..Source.. by The Crime Report

Back to the Future: The Influence of Criminal History on Risk Assessment

2-5-2015 National:

Abstract:
Evidence-based practices providing an empirical basis for predicting recidivism risk have become a primary focus across criminal justice decision points. Criminal history measures are the most common and heavily weighted factors in risk assessment tools, yet is such substantial reliance fully justified? The empirical and normative values placed on criminal history enjoy such commendation by criminal justice officials, practitioners, and the public that these practices are rarely questioned. This paper fills the gap by introducing and exploring various issues from legal, scientific, and pragmatic perspectives.

As a general rule, a common assumption is that past behavior dictates an individual’s likely future conduct. This axiom is often applied to criminal behavior, more specifically, in that prior offending is considered a primary driver to predict future recidivism.

Criminal justice officials have a long history of formally and informally incorporating risk judgments into a variety of criminal justice decisions, ranging from bail, sentencing, parole, supervisory conditions, and programming. A more contemporary addendum represents empirically informed risk assessment practices that integrate actuarial tools and/or structured professional judgments.

Various criminal history measures pervade these newer evidence-based practices as well. Instead of presuming the value and significance of prior crimes in judging future recidivism risk, this Article raises and critically analyzes certain unexpected consequences resulting from the significant reliance upon criminal history in risk assessment judgments.

Among the more novel issues addressed include: (1) creating a ratchet effect whereby the same criminal history event can be counted numerous times; (2) resulting in informal, three-strikes types of penalties; (3) counting nonadjudicated criminal behaviors and acquitted conduct; (4) proportionality of punishment; (5) disciplining hypothetical future crime; (6) punishing status; and (7) inadequately accounting for the age-crime curve.

In the end, criminal history has a role to play in future risk judgments, but these issues represent unanticipated outcomes that deserve attention. ..Source.. by Melissa Hamilton, University of Houston Law Center

A Better Path to Community Safety: Sex Offender Registration in California: Tiering Background Paper

March 2014 California:

The California Sex Offender Management Board has determined that there is a better path to the goal of achieving enhanced community safety through sex offender registration and notification practices and recommends that key changes be made to the state’s laws in these areas of sex offender management.

CASOMB first recommended in its 2010 report to the Legislature that revisions to California’s registration laws are needed and recently selected the issue as one which deserves increased focus and effort. The effectiveness of sex offender registration policies and practices has also been the subject of national focus recently, with a variety of jurisdictions addressing the importance of updating registration practices to reflect new research and evidence based approaches. Modifying registration practices will, CASOMB believes, improve public safety in California by focusing effort and resources on more dangerous offenders.

THE PRIMARY CONSIDERATIONS SUPPORTING THE NEED FOR
CHANGE TO THE SEX OFFENDER REGISTRATION SYSTEM IN CALIFORNIA

• 1. The goals of sex offender registration.
• 2. The assumptions behind the sex offender registry.
• 3. The current California picture.
• 4. What is known now that wasn’t known then?
• 5. What is still not known?
• 6. The problems with California’s current registration system.
• 7. The registry’s direct financial costs and cost-benefit ratio.
• 8. Indirect costs and other considerations.
• 9. Outline of the CASOMB proposal.

...Continued... by CSOMB

New Study Compares Sexual Predator Assessment Tools

June 2012:

The three most commonly used tools to assess the recidivism of a sexually violent predator (SVP) are the Static-99, the Minnesota Sex Offender Screening Tool, Revised (MnSOST-R), and the Psychopathy Checklist, Revised (PCL-R). Prison overcrowding and regulations that allow for hospitalization as an alternative to prison have resulted in the release of numerous SVPs. These individuals are granted release based in part on their risk for reoffending. These tools mentioned earlier are used in making this determination. Therefore, it is essential that they be fully evaluated to ensure that they are accurately assessing the potential recidivism rates of SVPs. However, few studies have compared these three tools in a clinical sample until now.

Cailey S. Miller of the Department of Mental Health Law and Policy at the University of South Florida believes that the risk to the community at large could be significantly reduced if SVPs were being effectively screened using the most valid and accurate tool possible. To determine which of these three tools can provide the most consistent results, Miller evaluated 315 SVPs using all three measures. She found that among all three tools, the Static-99 provided the most consistent results among raters, followed closely by the MnSOST-R.

Miller noted that the PCL-R was the least consistent in the results it provided; however, it had its strengths. Particularly, the PCL-R was very reliable at assessing antisocial behavior and lifestyle traits among the SVPs. However, Miller also found that all three instruments yielded field validity results that were far below those outlined in their respective manuals. Because SPVs pose significant traumatic, sexual, and psychological threats to communities, Miller believes that further exploration of these measures is necessary before they are universally used to grant release to offenders. Miller added that until then, “Evaluators can best inform this important legal decision-making process by familiarizing themselves with the strengths and limitations of those measures on which they rely and effectively communicating this information to the courts.” ..Source..

Predicting Reoffense for Community-Based Sexual Offenders: An Analysis of 30 Years of Data

May 2012:

Abstract:
This study contributes to the area of risk prediction by exploring whether the Static-99R is useful for predicting reoffense in community-based samples, and for noncontact offenders with and without identified victims. A total of 744 participants drawn from an outpatient sex offender treatment program in a large metropolitan area were followed for a period of up to 30 years. Multiple Cox Regressions were run; covariates included length of treatment, status in treatment, Static-99R items, and number of technical probation violations.

Overall, reoffending was an infrequent occurrence in this sample regardless of how it was defined, with sexual reoffenses identified in 13% of the sample and any criminal reoffense identified in 20% of the sample. Consistent with previous research, the Static-99R was a better predictor of sex-related reoffenses than of nonsexual reoffenses. However, in no case were more than a couple of the items significantly related to reoffending and these items differed depending on reoffense definition. ..Source.. by Rebecca E. Swinburne Romine; Michael H. Miner PhD; Dominic Poulin; S. Margretta Dwyer; Dianne Berg.

SVP risk tools show 'disappointing' reliability in real-world use

6-4-2012 National:

Rater agreement on three instruments commonly used to assess sex offenders' risk of recidivism is much lower in practice than reported in the tools' manuals, according to a new study out of Florida.

Faring most poorly was the Psychopathy Checklist (PCL-R). Correlations of scores between two evaluators hired by the same agency were in the low range. On average, psychologists differed by five points on the instrument, which has a score range of of zero to 40. In one case, two evaluators were apart by a whopping 24 points!

Agreement among evaluators was only moderate on the Static-99 and the MnSOST-R, two actuarial risk assessment instruments for which scoring is relatively more straightforward.

The study, published in the respected journal Psychological Assessment, was a collaboration between scholars from the Department of Mental Health Law and Policy at the University of South Florida and researchers with the Florida Department of Children and Families. It utilized archived records culled from the almost 35,000 individuals screened for possible Sexually Violent Predators (SVP) civil commitment in Florida between 1999 and 2009. The researchers located 315 cases in which the same individual was evaluated by separate clinicians who each administered both the PCL-R and at least one of the two actuarial measures within a short enough time frame to enable direct scoring comparisons.

It would be a mistake to lean too heavily on the results of a single isolated study. But the present study adds to a burgeoning body from several groups of independent researchers, all pointing to troubling problems with the accuracy of instruments designed to forecast risk of recidivism among sex offenders.

Related study: Psychopathy and sexual deviance not predictive

Collectively, the research has been especially critical of the ability of the highly prejudicial construct of psychopathy to add meaningfully to risk prediction in this high-stakes arena. Indeed, just this week another study has come out indicating that neither psychopathy scores nor sexual deviance measures improve on the accuracy provided by an actuarial instrument alone.

An especially interesting finding of that Canadian study is that reoffense rates were still below 12 percent over a 6-year followup period for even the most high-risk offenders -- those with high risk ratings on the Static-99R plus high levels of psychopathy and sexual deviance (as measured by phallometric testing). This makes it inappropriate to inflate risk estimates over and above those derived from Static-99R scores alone, the authors caution.

For the remainder of this article: by Karen Franklin, In the News

A Model of Static and Dynamic Sex Offender Risk Assessment

October 2011:

Abstract:
The purpose of the present study was to test models of combining static and dynamic risk measures that might predict sexual recidivism among adult male sex offenders better than any one type of measure alone. Study participants were 759 adult male sex offenders under correctional supervision in Vermont who were enrolled in community sex offender treatment between 2001 and 2007.

These offenders were assessed once using static measures (Static-99R, Static-2002R and VASOR) based on participants’ history at the date of placement in the community. A 22-item dynamic risk measure (SOTNPS) was used multiple times to assess participants, shortly after their entry into community treatment and approximately every six months thereafter. Analyses of SOTNPS scores resulted in the development of a new 16-item dynamic risk measure, the Sex Offender Treatment Intervention and Progress Scale (SOTIPS).

At fixed one- and three-year follow-up periods from participants’ initial, second, and third dynamic risk assessments, the SOTIPS and Static-99R, the static risk measure selected for further analysis in the present study, each independently showed moderate ability to rank order risk for sexual, violent, and any criminal recidivism and return to prison.

A logistic regression model that combined SOTIPS and Static-99R consistently predicted recidivism and outperformed either instrument alone when both instruments had similar predictive power. Participants who demonstrated treatment progress, as reflected by reductions in SOTIPS scores, showed lower rates of recidivism than those who did not. ..For the remainder of this research: by Robert J. McGrath, Michael P. Lasher, and Georgia F. Cumming, Vermont Department of Corrections (Grant Award Number: 2008-DD-BX-0013 National Institute of Justice)

---

...

Evaluating and Improving Risk Assessment Schemes for Sexual Recidivism: A Long-Term Follow-Up of Convicted Sexual Offenders

March 2007:

Executive Summary:
Risk assessment plays a central role in the management of sexual offenders in the criminal justice system, impacting every level of decision-making. The specification of which offenders should be included in state registration or community notification, the distribution of treatment opportunities in prison, the determination of parole eligibility, the allocation of resources in monitoring and supervising offenders in the community, and the ultimate decision to remove the most serious offenders from the community through civil commitment all involve critical choices involving risk. The serious consequences to potential victims, to those offenders who no longer pose a threat to the community, but are denied their civil liberties, and to critical community funds that may be squandered as the result of inaccurate decisions all demand the guidance of the most accurate actuarial algorithms available. Because decisions must be made—even doing nothing constitutes a decision—to choose not to employ the most accurate decision-making algorithms is to choose to pay the costly price of a suboptimal decision.

The aim of the present study was to evaluate and to improve the decision-making algorithms that have been generated to assess risk in sexual offenders. More specifically, it was the task of this research project to evaluate the extant actuarials in a sample of sexual offenders on whom long-term follow up were available. We assessed the comparative accuracy of the major risk instruments over time and over subsamples, explored their underlying factor structure, examined the accuracy of a new assessment protocol (SRA Need Assessment; Thornton, 2002), and explored the potential for generating improved predictive instruments.

In a prior study we (Knight, 1999; Prentky, Knight, & Lee, 1997) had followed 599 offenders who had been referred to the Massachusetts Treatment Center (MTC) for evaluation between 1959 and 1984. Of these 266 (Bridgewater Treatment [BT] sample) had been committed to MTC as “sexually dangerous” and subsequently released, and 333 (Bridgewater Observation [BO] sample) had been determined not to be sexually dangerous and returned to finish their sentences. Of the 333 BO offenders 200 constituted a matched sample (on age at evaluation, marital status, and number of prior crimes) and 200 were randomly sampled from the entire sample BO population evaluated. There was an overlap of 67 offenders selected by both the random and matched process. For all these offenders we had accessed and integrated four outcome record sources, including the Massachusetts Board of Probation records, the Massachusetts Parole Board records, the Massachusetts Treatment Center Authorized Absence Program records, and the Federal Bureau of Investigation (FBI) records.

In the present study we accessed the archival clinical files for these offenders and coded these records both on modern empirically-derived, mechanical actuarials that have been developed since 1997 for predicting sexual recidivism, including the RRASOR (Hanson, 1997), the Static-99 (Hanson & Thornton, 2000), the Static-2002 (Hanson & Thornton, 2003), the SORAG (Quinsey, Harris, Rice, & Cormier, 1998), the MnSOST-R (Epperson, Kaul, Huot, Hesselton, Alexander, & Goldman, 1998), and the Risk Matrix 2000 (Thornton et al., 2003), on two structured clinical guidelines, the SVR-20 (Boer, Hart, Kropp, & Webster, 1997) and the A-SOAP-II (Prentky & Righthand, 2003), and on a new experimental measure, the SRA Needs Assessment (Thornton, 2002). Offenders were also categorized in the Massachusetts Treatment Center Rapist (MTC:R3) and Child Molester (MTC:CM3) typologies and diagnosed according the DSM-IV Antisocial Personality Disorder and Conduct Disorder criteria. All codings and judgments were carried out without any access to or knowledge about the follow-up status of the offenders. ..For the remainder of this paper: by Raymond A. Knight, Ph.D. ; David Thornton, Ph.D.

OH- REPORT TO THE OHIO CRIMINAL SENTENCING COMMISSION: SEX OFFENDERS

January 2006

INTRODUCTION
The effective management of sex offenders has been an ongoing concern for policymakers nationally. In summer 2005, the Ohio Criminal Sentencing Commission created a Penalty Review Subcommittee to examine the research and current statutes in Ohio to determine if there was a need for recommendations to improve Ohio’s management of sex offenders. The purpose of this report is to provide information on sex offenders in Ohio’s prisons and discuss what works in effectively managing this population.

The first section provides a statistical snapshot of the offenders in 1999 at Ohio’s Sex Offender Risk Reduction Center (SORCC), sentencing information for calendar year 2003 offenders, and length of stay information for sex offenders released from the system in the past five years. The data given is designed to provide background information for the review of policy.

The second section of the report is a discussion of what works based on 15 years of research on sex offender assessment, treatment, and recidivism. Canadian and British researchers have been trying to identify the characteristics of the “sexual predator,” or the sex offender who is violent and causes the most harm to society. Colorado has had a sex offender management and containment approach since 1992 when the legislature created the Sex Offender Management Board as an oversight board for policy. The Board has completed several research studies on the effectiveness of treatment and management of sex offenders. This research indicates that with effective treatment and close supervision of offenders in the community, steps can be taken to reduce the likelihood they will reoffend. Finally, research on recidivism is presented together with an Ohio study which followed a 1989 cohort of offenders for 10 years.

The final section looks at sex offender registration and notification nationally and in Ohio. The report highlights current research on the effectiveness of SORN legislation in implementing the policy purposes and reducing recidivism of sex offenders in the community. Information on program implementation in Ohio has also been provided. ..more.. by Office of Criminal Justice Services, Executive Director Karhlton F. Moore

VA- Assessing Risk Among Sex Offenders in Virginia

2001 Virginia:

Executive Summary:
In 1999, the Virginia General Assembly requested the Virginia Criminal Sentencing Commission to develop a sex offender risk assessment instrument, based on the risk of re-offense, for integration into the state’s sentencing guidelines system. In accordance with Senate Joint Resolution (SJR) 333 of the 1999 General Assembly, the Commission embarked on an empirical study of recidivism among sex offenders convicted in the Commonwealth.

The Commission’s goal was to develop a reliable and valid predictive instrument, specific to the population of sex offenders in Virginia, that could be a valuable tool for the judiciary when sentencing sex offenders. If put in place, Virginia would be the first state in the nation to integrate sex offender risk assessment into sentencing guidelines.

Research Methodology
The Commission tracked 579 felony sex offenders who were released from incarceration (or sentenced to probation without an active term of incarceration) during fiscal years (FY) 1990 through 1993. Selecting offenders returning to the community from FY1990 to FY1993 allowed for a minimum five-year follow-up for all offenders in the sample, with some offenders followed for as long as ten years. On average, offenders in the Commission’s study were tracked for eight years. The offenders were selected in such a way that the overall sample reflects the characteristics of a random sample of sex offenders sentenced in Virginia’s circuit courts in calendar years (CY) 1996 and 1997. This design enables the Commission to generalize the results of the study to the population of sex offenders sentenced in circuit courts in the Commonwealth.

Automated data was supplemented through manual data collection. Through examination of narrative accounts found in pre/post-sentence investigation (PSI) reports, rich contextual detail of the sex offenses committed by offenders in the sample was gathered. Criminal history “rap sheets” from the Virginia Criminal Information Network (VCIN) system maintained by the Virginia State Police and from the FBI’s Central Criminal Records Exchange (CCRE) system provided recidivism data and supplemented prior record information.

Measuring Recidivism
There are many barriers to accurate measurement of recidivism among sex offenders, including reluctance to report sexual victimization and evidentiary problems resulting in offenders not being charged or convicted of their crimes. Victims and witnesses may refuse to come forward to testify, particularly when the victim is young. These and other obstacles hinder the prosecution of sex offense cases and often mean that charges must be dropped or reduced in a plea agreement. In order to avoid the underestimation of recidivism that is inherent with measurement based solely on reconviction, the Commission elected to define recidivism using official records of arrests. The Commission believes that measuring recidivism by a new arrest more closely approximates the true rate of re-offense behavior among sex offenders. To the extent that sex offenders go on to commit other types of violent crimes, re-arrests for new sex offenses will underestimate the predatory nature of these offenders. The Commission, therefore, chose as its operational definition of recidivism a new arrest for a sex offense or any other crime against the person.


Treatment of Sex Offenders
SJR 333 requests the Commission to consider the impact of treatment interventions on the reduction of recidivism among this particular population of offenders. The Commission, however, determined that assessing the effectiveness of post-conviction treatment services among offenders in the study sample would be extremely difficult.

In 1992, the Joint Legislative Audit and Review Commission (JLARC) determined that, during the time in which the offenders under study were incarcerated, “the Department of Corrections had not promulgated any standards to govern the development of treatment programs in the prisons and field units” (p. iii). JLARC found no agency specific requirements for the service providers, no minimum qualifications for counselors conducting group therapy and no guidelines outlining the basic elements of therapeutic counseling (p. iv-v). Moreover, only half (53%) of imprisoned sex offenders received any treatment services prior to reaching their first parole eligibility date (JLARC 1992, p. iv). Of those receiving treatment services when they become eligible for parole, a large share (40%) were provided only sex offender education programming, and not sex offender therapy. Furthermore, little consistent documentation about participation in prison-based sex offender treatment programs was available in files at the headquarters of the Department of Corrections.

Given these serious limitations, the Commission concluded that the impact of post-conviction treatment and its effect on rates of recidivism among sex offenders returned to the community from FY1990 through FY1993 could not be accurately assessed as part of the current study. Although the impact of specialized sex offender treatment provided after conviction was not examined, the Commission analyzed available automated data indicating whether or not the offender had received some type of mental health treatment or participated in an alcohol or drug treatment program prior to committing the offense under study. A review of literature on the effectiveness of sex offender treatment is provided in this document.

Risk Assessment Findings
In risk assessment research, the characteristics, criminal histories and patterns of recidivism among offenders are carefully analyzed. Factors proven statistically significant (i.e., those with a known level of success) in predicting recidivism can be assembled on a risk assessment worksheet, with scores determined by the relative importance of the factors in the statistical model.

The risk assessment instrument developed by the Commission reflects the characteristics and recidivism patterns of the population of felony sex offenders convicted and sentenced in Virginia. The risk assessment instrument proposed by the Commission can be found in Figure 26 on page 56 of this report. Relevant findings are presented below.

For the remainder of the study: by Virginia Criminal Sentencing Commission

Note: Nice Overview of Several SO Studies

Two Important Charts:
Location of Offenses & Victim / Offender Relationship