One of the most frequently cited and dispiriting statistics about the American criminal justice system is that more than half of state prisoners end up returning to prison within five years of their release.* These numbers come from a study conducted by the federal government’s Bureau of Justice Statistics, in which researchers tracked about 400,000 people from around the country who were released from state prisons in 2005. The strong implication of the findings is that people who are incarcerated are extremely likely to reoffend once they’re free and that most of them spend their lives in and out of correctional facilities.
But what if the BJS’s findings have been fundamentally misunderstood? That’s the provocative contention of a recent paper published in the journal Crime & Delinquency, the title of which is “Following Incarceration, Most Released Offenders Never Return to Prison.”
The paper, which was produced by researchers at the Cambridge, Massachusetts–based public policy firm Abt Associates and circulated online this week by criminal justice experts, argues that the conventional wisdom about recidivism in America is flatly wrong. In reality, the authors of the paper report, 2 out of 3 people who serve time in prison never come back, and only 11 percent come back multiple times.
The reason for the shocking discrepancy between these new findings and those of the BJS, according to Abt’s William Rhodes, is that the BJS used a sample population in which repeat offenders were vastly overrepresented.
I called Rhodes to ask him about why this happened and how he and his co-authors avoided the same problem in their analysis. His explanation for why the recidivism problem is not nearly as bad as many of us have believed is below; our conversation has been lightly edited and condensed.
Let’s start with the conventional wisdom on recidivism in the U.S. What is it, and where did it come from?
The conventional wisdom is that there’s a very high rate of recidivism, where recidivism is defined as being arrested for a new crime or having your community supervision status revoked for a technical violation.
I know the Bureau of Justice Statistics has collected statistics on recidivism at least twice, maybe three times, and what they do is start with a sample of offenders who are released from prison during a given year, then match those release records with criminal history records to determine who recidivates. Then they compute their statistics—the rate at which the released offenders are arrested for new crimes and the rate at which they’re readmitted to prison—by observing the individuals in their sample over a period of some years. They’re not controversial statistics. There’s no manipulation that goes on. It’s purely tabulation.
So the way it works is they choose a year and track a cohort of people in their sample and see who comes back? It seems pretty straightforward.
That’s exactly right.
So what’s wrong with their results?
It is difficult to explain to a nonstatistician. I try to use an analogy: Suppose that I were asked to describe a population of people who go to shopping malls. What I might do is go to the mall and perform an “intercept survey”—that is, I’d randomly select people who are entering the mall and ask them about themselves—record their age, sex, race, and frequency of visiting the mall. The problem is, I’d probably do that over a pretty short period of time, like a week. So I’d get a lot of people who are frequent mall visitors and fewer people who aren’t. You know, if you go to a mall you’ll see an elderly population who go daily, to exercise by walking through the mall. You’ll also see a number of people who simply like malls, and maybe they go weekly. Or you’ll find, occasionally, people like me, who go about once a year when they need to buy a washing machine or something. If you did a simple tabulation of all the people you intercepted during a week you’d get a large proportion of frequent mall visitors. And they wouldn’t be representative of people who visit malls—they’d be representative of frequent mall visitors.
And the same thing is happening with the Bureau of Justice Statistics when they take a sample of people who have been released from prison during a given year.
Right. They’re not attempting to be misleading. What they’re reporting is true: If you take people who are released from prison during a given year, here’s the rate at which they’ll return. But it gets translated in people’s heads as, “Here’s what happens to offenders in general.”
In truth what you have is two groups of offenders: those who repeatedly do crimes and accumulate in prisons because they get recaptured, reconvicted, and resentenced; and those who are much lower risk, and most of them will go to prison once and not come back.
So the problem with taking a snapshot of a particular year, the way the BJS has done it, is you’re more likely to have people in your sample who come back a lot than you are to have people who don’t come back at all.
That’s exactly right, yeah.
What data is your study based on?
At Abt Associates we assemble data into something called the National Corrections Reporting Program. It records prison terms for offenders across almost all of the states. For a large number of states, that data goes back to 2000. So we can observe when somebody enters and when somebody exits prison, and that allows us to look at individual offenders and say, “Given that they’ve been incarcerated at least once, how frequently do they come back?” So you’re looking at a large number of offenders, over a nearly 15-year period, and what you find is that most of those offenders do not come back. They’re incarcerated, they serve their term, they don’t return.
So your data set contains information at the individual level? You know when a specific person went in and when he got out?
That’s correct. If you were in the dataset, we would track you. We probably wouldn’t have your name, but we’d have an identification code that the state would issue you as an inmate.
So what do you have to do to correct for the overrepresentation of repeat offenders in the dataset?
You weight them differently. It’s not arbitrary of course—the weighting is done so that you have an appropriate representation of all offenders rather than an overrepresentation of high-rate offenders. In order to get the right weights, you have to be able to observe a long period—the 15 years we look at.
So the reason you’re looking at the stretch of time, rather than just one year, is it gives you what you need to know in order to weight specific individuals the right amount.
So looking at your findings and the findings that usually get reported, it’s almost like we have these two very different realities. We have one that says “most people come back” and we have one that says “most people don’t come back.” Given that, as you said, there’s nothing misleading about what the BJS was doing in their analysis, what is going on? And which one should we look at if what we’re interested in is knowing how likely it is that someone who has served time in an American prison will return?
What’s important is being clear about what question you’re trying to answer. If your question is, “Of all the people who go to prison, what’s the rate at which they come back?” then our calculations are better. But if you wanted to ask a question about a specific release cohort—about people who are released during a given year—and how frequently they come back, then the other methodology is the appropriate one. But they’re questions about two different populations of people. The first one is the population of offenders in general.
So what are the BJS numbers good for?
Well, there are reasons for using data like that. You might do it if you wanted to evaluate whether a program you introduced in prison reduced recidivism. So you’d want to look at a particular cohort that was released during a particular year and judge whether the treatment you introduced was effective or not.
But if you want to look at how offenders actually interact with the criminal justice system, then the methods we propose are more appropriate.
My understanding of the lives of people who go to prison was very much colored by the notion that they tend to be incarcerated over and over—that they come out of prison and they have a very small chance of staying free. What you’re saying is that’s only really true for a certain subset of the population of people who are incarcerated.
Yes, that’s correct. Most people really do not return to prison. They’re not caught in what we call the cycle of incarceration. They don’t churn, to use one of the popular words. But some do.
Are there policy implications from this that you’ve thought about?
Yeah, I think there are. It would take more careful study, but others have pointed out that there are very low-level offenders who manage to readjust, and you ought to focus the rehabilitation resources you have on those individuals who are high-risk offenders. They’re the ones who are going to benefit most from treatment—or, I should say, society’s going to benefit most from treating them. The problem, of course, is identifying them. That’s why criminologists have attempted to develop risk assessment tools, to identify the high-risk offenders and treat them, while almost letting the others recover by themselves.
Correction, Nov. 2, 2015: This article originally misstated that a Bureau of Justice Statistics study on recidivism found that 68 percent of state prisoners ended up back behind bars within three years of their release, and about 75 percent came back within five. These numbers referred to rates of re-arrest, not re-imprisonment. The BJS study found that about 50 and 55 percent of state prisoners returned to prison within three and five years, respectively. (Return.)