Lies, damned lies, and recidivism statistics

January 2008:

Thanks to this post at Sex Crimes, I see that the Numbers Guy at the Wall Street Journal has this effective post asking the important question "How Likely Are Sex Offenders to Repeat Their Crimes?" and this related print story about the challenges of getting accurate data on sex offender recidivism.  Here are snippets of the blog post from the Numbers Guy:

In debates over laws monitoring released sex offenders, it’s common to hear claims that they’re sure to commit more sex crimes.... But as my print column this week points out, the numbers don’t bear this out.  Recidivism rates vary widely depending on which crimes are counted, the timeframe of the studies, and whether repeat offenses are defined by convictions, arrests, or self-reporting.  But even the author of a widely published report suggesting a recidivism rate of 52%, Wisconsin psychologist Dennis Doren, told me of the notion that all sex criminals are likely to re-offend, “There is no research support for that view, period.” ...

The conventional wisdom on sex-crime recidivism, coupled with high-profile sex crimes against children, has helped spur the spate of registry and neighbor-notification laws, even before they could be properly studied for their impact on recidivism rates. Several researchers, including Dr. Doren, say that residency-restriction laws may be counterproductive. Such a constraint “drives them out of their community, and leads to a lack of stability,” said Karen J. Terry, a criminologist at John Jay College in New York. “Those are some of the underlying conditions that caused them to abuse in the first place.” A consensus on how to measure recidivism, and determine its baseline rate, would help evaluate such laws.

This research is expensive and long-term follow-ups are, by definition, slow to produce results.  Even if we were to know whether rates have declined in recent years, it would be difficult to isolate the cause.  Dr. Doren proposes several alternate explanations for his perception that rates have declined in recent years, including better and more frequent treatment, and closer monitoring.

Critically, the challenging issues spotted by the Numbers Guy attend not only to sex offender recidivism.  The US Sentencing Commission did some very interesting work on recidivism a few years ago (see here and here and here), which found among other things that the recidivism predictor used by the the US Parole Commission two decades ago "is a statistically better recidivism risk prediction instrument than" the criminal history categories incorporated by the Sentencing Commission into the sentencing guidelines.

What's most disconcerting, however, is the common reality that even perfect crime rate studies and data about recidivism cannot alone significantly alter the public viewpoints and political debates.  As we often see in the context of the death penalty, statistical realities and empirical debates often serve —perhaps sometimes just unconsciously —  as cover for the expression of other normative concerns.  Though I sincerely hope lots of smart folks will continue crunching public safety numbers, policy advocates need to invest more time in thinking about how to effectively operationalize solid data insights into sound sentencing reforms.

..Source.. by Sentencing Law and Policy

How Likely Are Sex Offenders to Repeat Their Crimes?

January 2008:

In debates over laws monitoring released sex offenders, it’s common to hear claims that they’re sure to commit more sex crimes. “‘What we’re up against is the kind of criminal who, just as soon as he gets out of jail, will immediately commit this crime again at least 90 percent of the time,” a California legislator told the New York Times in 1996. (Other examples of such rhetoric are collected here.) Fox News — like the Wall Street Journal owned by News Corp. — said of child molesters in 2005, “Not only are they almost certain to continue sexually abusing children, but some eventually kill their young victims.”

But as my print column this week points out, the numbers don’t bear this out. Recidivism rates vary widely depending on which crimes are counted, the timeframe of the studies, and whether repeat offenses are defined by convictions, arrests, or self-reporting. But even the author of a widely published report suggesting a recidivism rate of 52%, Wisconsin psychologist Dennis Doren, told me of the notion that all sex criminals are likely to re-offend, “There is no research support for that view, period.” Dr. Doren, evaluation director at the Sand Ridge Secure Treatment Center in Mauston, Wisc., added, “You’re not talking to a bleeding-heart kind of guy here.”

Yet incorporating convicted sex offenders’ undetected crimes can lead to higher numbers, such as one controversial Canadian study that found long-term recidivism could be as high as 88.3%. (It was debated in the Canadian Journal of Criminology and Criminal Justice in 2006.) Critics say that rate was artificially increased by the study’s design.

This isn’t just an academic exercise. The conventional wisdom on sex-crime recidivism, coupled with high-profile sex crimes against children, has helped spur the spate of registry and neighbor-notification laws, even before they could be properly studied for their impact on recidivism rates. Several researchers, including Dr. Doren, say that residency-restriction laws may be counterproductive. Such a constraint “drives them out of their community, and leads to a lack of stability,” said Karen J. Terry, a criminologist at John Jay College in New York. “Those are some of the underlying conditions that caused them to abuse in the first place.” A consensus on how to measure recidivism, and determine its baseline rate, would help evaluate such laws.

This research is expensive and long-term follow-ups are, by definition, slow to produce results. Even if we were to know whether rates have declined in recent years, it would be difficult to isolate the cause. Dr. Doren proposes several alternate explanations for his perception that rates have declined in recent years, including better and more frequent treatment, and closer monitoring.

Meanwhile, the existing research raises tough questions about the relative danger child molesters pose to society. Their likelihood of being convicted for a crime after release is much lower than average for all criminals released from prison, and even for all sex offenders, at least in the short term, as measured by a Bureau of Justice Statistics study and others. Yet their crimes, when they do repeat child abuse, are unusually harmful, and their victims particularly vulnerable. Does that justify the closer monitoring of child molesters after release, compared with other criminals? Dr. Doren isn’t sure, pointing out, for example, that convicted rapists are more likely to re-offend in the years immediately after release, and more likely to commit other violent crimes. “If we’re concerned about violence generically, it’s rapists we should be concerned about” in the short term, he said.

What do you think? What is the best way to measure recidivism rates, and what should be measured? Do the numbers justify registries and neighbor notification? Please let me know in the comments.

Further reading: Other newspapers previously have pointed out that measured recidivism rates appear to contradict conventional wisdom. Some states are attempting to fine-tune their post-release plans for sex offenders based on differing rates of recividism; here’s an article about the plan in Texas. Illustrating the difficulty of measuring the effect of interventions on recidivism rates, a Canadian study found that sex offenders who completed treatment were far less likely to re-offend — but that may not mean the treatment itself was successful. Instead, it might just demonstrate that a willingness to complete treatment is an indicator of other factors that diminish the likelihood of repeating a sexual crime. ..Source.. by Carl Bialik, Wall Street Journal

Sex Offender Recidivism: A Simple Question

2004:

Abstract

This [Meta] study examines sexual recidivism, as expressed by new charges or convictions for sexual offences, using the data from 10 follow-up studies of adult male sexual offenders (combined sample of 4,724). Results indicated that most sexual offenders do not re-offend sexually, that first-time sexual offenders are significantly less likely to sexually re-offend than those with previous sexual convictions, and that offenders over the age of 50 are less likely to re-offend than younger offenders.

In addition, it was found that the longer offenders remained offence-free in the community the less likely they are to re-offend sexually. Data shows that rapists, incest offenders, "girl-victim" child molesters, and "boy-victim" child molesters recidivate at significantly different rates.

These results challenge some commonly held beliefs about sexual recidivism and have implications for policies designed to manage the risk posed by convicted sexual offenders. ..Source.. by Andrew J. R. Harris and R. Karl Hanson, Public Safety and Emergency Preparedness Canada

Quote from the study:

A Simple Question

The basic question about sexual offender recidivism is usually phrased along the following lines: "what percentage of sexual offenders commit another sexual offence once they've been released from prison?" This question is not as easy to answer as one might believe.

First, we must define "recidivism". In some studies, recidivism is defined as a reconviction for a sexual offence (e.g., Hanson, Scott & Steffy, 1995). In other studies, recidivism includes all offenders who were charged with a new sexual offence, whether or not they were convicted (e.g., Song & Lieb, 1995). Including charges along with convictions should, of course, lead to higher estimates of recidivism (Prentky, Lee, Knight & Cerce, 1997). Other studies have used expanded definitions of sexual recidivism that include informal reports to child protection agencies, self-report, violations of conditional release conditions, and simply being questioned by police (e.g., Marshall & Barbaree, 1988).

All else being equal, the estimated recidivism rate should increase with each expansion of the definition; the broader the definition, the larger the recidivism estimate should appear. Consequently, it is important to specify the recidivism criteria in any recidivism estimate (e.g., "what percentage of sexual offenders are either charged with, or convicted of, another sexual offence once they've been released from prison?")

Another factor to consider is the length of the follow-up period. As the follow-up period increases, the cumulative number of recidivists can only increase. It is important to remember, however, that an increase in the number of recidivists is not the same as an increase in the yearly rate of recidivism. For all crimes (and almost all behaviours) the likelihood that the behaviour will reappear decreases the longer the person has abstained from that behaviour. The recidivism rate within the first two years after release from prison is much higher than the recidivism rate between years 10 and 12 after release from prison. Consequently, any estimate of sexual re-offending must be "time-defined" or "time limited" (e.g., "over the first five years, post-release from prison, what percentage of sexual offenders are either charged with, or convicted of, another sexual offence?")

A third factor to consider is the diversity among sexual offenders. We know that incest offenders recidivate at a significantly lower rate than offenders who target victims outside the family (Hanson & Bussière, 1998). We also know that child molesters with male victims recidivate at a significantly higher rate than child molesters that only have girl victims (Hanson & Bussière, 1998). By considering the type of sexual offender, our simple question becomes, once again, more complex: (e.g., "over the first five years, post-release from prison, what percentage of child molesters with male victims are either charged with, or convicted of, another sexual offence?")

Many sexual offences are never reported to police; this is the same for all violent offences except murder. Our best estimates of unreported sexual offending come from victimization studies. In a typical study a random sample of people are telephoned and asked if they have been a victim of a crime within the last year. One recent victimization study found that there were approximately half a million sexual assaults (499,000) committed in Canada in 1999 (Besserer & Trainor, 2000). Although reports to police of violent and sexual crimes were steadily declining in Canada between the years 1993 and 1999; the years 2000 and 2001 saw 1% increases in violent and sexual crimes (Savoie, 2002). Sexual victimization rates based upon victimization surveys appear to have remained basically unchanged across this same time period (Besserer & Trainor, 2000). The Besserer and Trainor (2000) study showed that sexual assault had the highest percentage of incidents that were not reported to police (78%). When respondents were asked why they did not report sexual victimization to the police, 59% of the respondents stated that the "incident was not important enough" to report. Consequently, readers may wonder what counts as a sexual assault.

The Besserer and Trainor (2000) victimization study used a very broad definition of sexual assault. They counted all attempts at forced sexual activity, all unwanted sexual touching, grabbing, kissing, and fondling, as well as threats of sexual assault (Jennifer Tuffs, personal communication, January 15, 2003). Their broad definition undoubtedly included some behaviours that do not conform to the popular image of a sexual offence.