The Promise (and Perils) of Predicting Sex Crimes


Attorney-General Eric Holder's August 1 speech criticizing the use of risk assessment in sentencing decisions may not lever the issue to the top of the policy agenda. But a new paper could revive the debate about the effectiveness of risk tools in evaluating the chances of recidivism among those convicted of sex crimes.

A forthcoming article in the Arizona State Law Journal argues that state criminal justice systems which use risk assessment tools may overestimate sex offenders' likelihood of committing another crime. That message may complicate the efforts of those who advocate reform of sex offender policies. A key goal of reformers is to have states use actuarial risk assessments to classify offenders, instead of basing risk levels on their crime of conviction, as required by the 2006 federal Adam Walsh Act.

“Actuarial” risk assessments are designed to let state criminal justice systems evaluate risk as car insurance companies do. A list of factors that correlate with recidivism is used to group offenders into categories. Offenders with factors that correlate with higher reoffense rates are judged at greater risk; those with fewer factors are put in a lower tier.

Such assessments are increasingly used in release decisions for all types of offenders. A 2008 survey by the Association of Paroling Authorities International found that 32 out of 37 responding states used risk assessment tools to help determine conditions of parole or probation.

For those convicted of sex crimes, the stakes in risk assessment are high. Those identified as likely to commit a new offense appear on public state sex offender registries. In many states and cities, they're also banned from living near schools, parks, or daycares.

Research shows that actuarial risk assessments perform better than what they replaced—relying on experts to use their experience to make unstructured clinical judgments. Studies done in the late 1990s concluded that professionals don't perform much better than chance in predicting recidivism.

But criminal law professor Melissa Hamilton, the Law Journal article's author, sees flaws in the leading actuarial instruments. These instruments, she writes, often look only at historical, or “static,” factors in an offender's background, particularly their number and type of offenses.

Moreover, they don't take into account more recent “dynamic” factors that might increase an offender's chances of staying clear of future crime—success in a treatment program, abstention from alcohol and drugs, or a strong support network, for example.

At least two previous studies also concluded that the inability of static instruments to take into account changes in offenders' lives weakened their validity, according to a December 2012 meta-analysis of studies on sex offender risk assessment tools in the journal Clinical Psychology Review.

(The Static-99 is the most popular static risk tool. The lead developer was unavailable to respond to requests for comment, and a member of the Static-99 Clearinghouse advisory board told The Crime Report that the board also couldn't comment.)

Another Hamilton criticism focuses on the way assessment tools are used, rather than on the tools themselves. Some states give evaluators the option of “overriding” a tool's score if their professional judgment indicates that the offender should have been scored higher. Hamilton, a former corrections and police officer, told The Crime Report that evaluators are more likely to throw out low rather than high scores, in part because of widely held assumptions that all sex offenders are at high risk to recidivate.

Her critique could complicate the efforts of those who advocate using individual risk assessments to make tiering decisions about sex offenders instead of the offense-based categories required by the Adam Walsh Act. (The Walsh Act had been implemented by only 17 states as of April.) The Association for the Treatment of Sexual Abusers, for example, has advocated that sex offender tiering decisions be made using “empirically derived risk assessments rather than the unproven offense-based categories proposed in the Adam Walsh Act.”

That may sound like a technical debate, but the implications are enormous.

Offense-based tiering results in many more of those on state sex offender registries being classed as high risk. A 2006 article in the journal Criminal Justice and Behavior, for example, found that after Ohio and Oklahoma adopted the Walsh Act's offense-based tiering system, most registrants were shifted from lower risk to higher risk tiers.

Researchers have found problems with the Walsh Act's tiering system. A November 2012 report released by the Justice Department concluded that it performs poorly in predicting recidivism. The authors concluded that a system based on more empirical data needs to be developed.

If Hamilton's paper provokes questions about whether actuarial risk assessment works, it's unclear what would replace it. One alternative is to use newer risk assessment tools that consider dynamic factors. The December 2012 Clinical Psychology Review meta-analysis looked at 43 studies of 15 risk assessment tools to see which tools most accurately predicted recidivism for adult male offenders. The researchers concluded that the two tools found to perform best in predicting risk both incorporated dynamic factors. But that conclusion came with a caveat: since they're newer, they haven't undergone as many evaluations as the older static tools.

And some states use static instruments as only part of a more extensive evaluation. In Arkansas, for example, staff conduct individual assessments of those convicted of sex crimes before release or when an offender moves in from out of state.

They collect data about any previous offenses that the person was convicted of or charged with, conduct face-to-face interviews, and can require the person to take a polygraph. Subjects are scored on two risk assessment instruments—the Static-99, and another that considers dynamic factors. All of that is used to determine the offender's risk level for purposes of community notification, says Sheri Flynn, administrator of sex offender screening and risk assessment at the Arkansas Department of Correction.

“The Static-99 is an excellent instrument and it's by far one of the best researched and well supported,” says Flynn. “But you cannot ignore in our opinion those dynamic factors. If you just look at the number of times people have been convicted, you're missing the entire boat.”

A system like that may avoid overreliance on a single tool to perform what's after all a perhaps- unachievable task: predicting human behavior. A February article in Federal Sentencing Reporter makes a somewhat grim observation about the future of risk assessment, observing that “it is possible that each instrument is reaching a 'natural limit' to predictive utility, beyond which it cannot improve.”

Steven Yoder writes about criminal justice, immigration, and other domestic policy issues. His work has appeared in Salon, The Fiscal Times, The American Prospect, and elsewhere, and he's working on a book about the impact of sex offender registries on registrants' families. Online at @syodertweets. He welcomes comments from readers.

Comments are closed.