What the Recidivism Numbers Really Say About Second Chances

Every side of the prison debate has a favorite number, and every side is a little embarrassed by the other side's number. Reformers say most people who leave prison can rebuild a life. Hard-liners say most of them go straight back to crime. Here is the awkward part: the government's own data lets each camp quote a true figure, and neither camp likes the full table.

So let us read the full table. It comes from the Bureau of Justice Statistics, the statistical arm of the Justice Department, which followed prisoners released from the state prisons of 30 states in 2005 and tracked their arrests for nine years. The report, published in 2018, is a very large federal study of released prisoners. It deserves to be read slowly, with a pencil, by people who want to argue about parole and reentry honestly.

Define the terms before you fight about them

A number is only as good as its definition. Here are the terms this post uses.

Recidivism means a former prisoner returns to criminal behavior. That is the idea. The measurement is another matter, and researchers use at least three yardsticks: re-arrest, re-conviction and re-incarceration. Each answers a different question.

Arrest-based recidivism, the yardstick in the BJS study, counts a released prisoner as a recidivist the first time he is arrested for any new offense, in any state, after release. An arrest is an accusation by a police officer. It is not a verdict. It is, however, the earliest and most complete signal that something has gone wrong, and it is the only one that does not depend on how fast a court docket moves.

Cohort means the group followed. Here the cohort was the 2005 releases from 30 states, about 401,000 people. Those 30 states accounted for 77 percent of all state prisoners released that year, which is why the study carries weight.

Base rate means how often something happens across the whole group. Almost every bad argument about recidivism confuses the base rate with the behavior of the typical person, and we will see how in a moment.

The numbers, read straight

The 2018 report, titled 2018 Update on Prisoner Recidivism: A 9-Year Follow-up Period (2005-2014), reported cumulative arrest rates for the 2005 cohort. Here is the headline series.

Read the table the way a hard-liner would and you get a grim picture: more than four in five released prisoners were arrested again within nine years. Read it the way a reformer would and you notice something else. The curve is steepest at the beginning. Forty-four percent of released prisoners had been arrested by the end of year one, more than half of the eventual 83 percent, and the curve flattens afterward. From year six to year nine the cumulative rate rises only four points. The risk is front-loaded.

That shape matters more than the headline. If risk is concentrated in the first months after release, then the first months are where policy should spend its effort: housing, a job lead, a supervision officer who answers the phone, a way to get a state identification card before the money runs out. Parole and reentry programs that arrive in month eleven are a day late.

The same report found two other facts worth holding in mind. First, the arrest rate fell with age at release. Prisoners released at forty and older were arrested at lower rates than those released at twenty-four or younger. Crime is a young man's occupation, and it is one that most men eventually leave. Second, the study counted arrests, and the cohort produced a very large number of them: on the order of two million arrests over the nine years, for all kinds of offenses, from drug and public-order charges to violent crimes. That is a number to hold against the reformer's cheerful tone, and it is the kind of number that turns a ribbon-cutting into a hearing.

What the arrest number does and does not measure

Now the hard part, because a number that is quoted without its limits is a number that misleads. The nine-year figure has five limits, and honest people on both sides should state them.

1. An arrest is not a conviction. Some arrests end in dismissal. The BJS report is explicit that it counts arrests, and the share of arrests that become convictions is lower than the share of releasees who were arrested. A reformer is entitled to say that 83 percent re-arrested does not mean 83 percent re-offended.
2. An arrest is not a trivial event either. A hard-liner is entitled to answer that arrest data undercount crime. Most crimes never produce an arrest at all. Someone who is arrested once may have offended many times. If you want a floor on recidivism, re-arrest is closer to the floor than the ceiling.
3. "Any offense" is broad. The measure counts a probation violation that leads to an arrest the same way it counts an armed robbery, unless the report breaks the offenses out. A person re-arrested for a drug possession charge and a person re-arrested for aggravated assault both appear in the 83 percent. The mix of offenses matters, and the report does break them out; readers should look at the offense tables before they quote the top line.
4. The cohort is state prisoners. Federal prisoners, jail inmates and people on probation who never went to prison are different populations with lower rates. A figure for state prisoners is not a figure for "offenders."
5. The measure is a count of people, not of crimes. One arrest counts the same as twelve. That leads to the most important point in this post.

The arithmetic of a few repeat offenders

Here is a hypothetical, labeled as one, because it shows why base rates deceive. Picture a cohort of 1,000 released prisoners. Suppose, for the sake of the arithmetic, that 200 of them are high-rate offenders who are each arrested four times over a period, and that another 300 are arrested exactly once. The remaining 500 are never arrested again.

• The share arrested at least once is (200 + 300) divided by 1,000, which is 50 percent.
• The total arrests are (200 times 4) plus (300 times 1), which is 800 plus 300, or 1,100 arrests.
• The 200 repeat offenders are 20 percent of the cohort and account for 800 of the 1,100 arrests, which is about 73 percent.

Change nothing about the other 800 people. Now suppose a policy keeps those 200 repeat offenders from offending, through incarceration or through intense supervision. The share arrested at least once drops from 50 percent to 30 percent, and the total arrests drop from 1,100 to 300. The headline recidivism rate fell by twenty points, but the number of crimes prevented fell by nearly three quarters.

That is a hypothetical, not a finding. But the pattern in it is one of the most replicated findings in criminology: a small share of offenders produces a large share of offenses. The BJS data bear it out in the plain way that the arrest counts are skewed. It follows that the headline "percent who return" is the wrong target for policy. A program can move that number very little and still prevent a great deal of crime if it finds the few who do the most damage. And a program can move it a great deal for people who were never going to do much harm.

This cuts both ways, and that is why I call the numbers sobering for both camps. For the reformer: do not claim that most releasees are harmless. Four in five were arrested again, and the repeat offenders among them are real, and their victims are real. For the hard-liner: do not claim that the arrest rate proves that nobody changes. In the hypothetical above, half the cohort never came back. In the real data, the arrest curve flattens, and the likelihood of a new arrest falls with every year of age and every year of stability.

The steelman: second chances are the point

Let us state the reformer's best case in words its defenders would accept. It runs like this.

Nearly all state prisoners will be released at some point. The logic is simple: sentences end. The question is never whether these men come home, only what shape they are in when they do. A prison system that releases a man with a bus ticket, a small sum of gate money and a record that bars him from half the jobs in his county has designed a failure. If a returning prisoner cannot find work or a place to sleep, we should not be surprised when he falls back on the only trade he has been practicing. Second chances are therefore not charity. They are public safety by another route.

That argument is stronger than its critics admit, and it has deep roots on the right. The Judeo-Christian tradition does not treat a man's worst act as his permanent definition. It asks for repentance and it expects restitution, and it takes both seriously. A conservative who believes in redemption and in personal responsibility is not choosing between them. He is holding both, which is harder.

The rebuttal: second chances are earned, and the public comes first

Now the answer, in numbered points.

1. Public safety is the first duty of the criminal law. The Preamble lists "insure domestic Tranquility" among the reasons for the Constitution. A reentry program is a means. If it releases men who are arrested within a year at a rate of 44 percent, the first question is not whether the program is kind but whether it is working.
2. The data support targeting, not blanket leniency. Because a small share of offenders commits a large share of crimes, a policy that treats every releasee as equally likely to reform wastes resources on those who need little and under-supervises those who need a great deal. Risk assessment, the sorting of releasees by their likelihood of reoffending, is the honest way to spend a limited budget. The First Step Act of 2018 built a federal version of this idea, a point we take up in our post on How Earned Time Credits Work in Federal Prison, and Who Is Left Out.
3. Second chances should be earned. A conservative case for reentry asks something of the released man: work, sobriety, restitution, compliance with the conditions of release. A parole system that frees on a calendar date with no demonstrated change is not a second chance. It is a coin flip played with other people's safety, and the people flipping it are never the people who live next door.
4. Victims are part of the ledger. Every arrest in those 1.99 million-or-so entries has someone attached to it. A reentry policy that talks only about the returning prisoner has left half the people out of the room. We discuss how a victim should be treated in What to Do if You Are a Crime Victim, Your Rights Step by Step.

What the numbers imply for parole

Put the premises together and several conclusions follow, and I will state them as opinions, because they are.

Front-load supervision. If 44 percent of releasees are arrested within the first year and 68 percent within three, supervision resources belong in those windows. A supervision officer with a caseload so large that he sees a releasee for ten minutes a month is not supervising. He is bookkeeping, and the books are being kept on a house that is on fire.

Use risk, not mood. Parole boards should lean on validated risk instruments and on the record of conduct inside, and should explain their reasoning in writing so that victims, prosecutors and the public can see it. Discretion is fine. Unexplained discretion is a problem in a republic.

Test what you fund. The best evidence on reentry programs is mixed, and honesty requires saying so. Some job-training and supervision programs show real effects in careful evaluations, and others show none. A conservative who distrusts government programs should be the first to demand evaluation, and to cut what fails. The faith-based and charitable programs that have long worked with prisoners deserve the same scrutiny as any government one. We look at one remarkable example in Chuck Colson Went From Watergate to Prison to a Ministry Behind Bars.

Do not mistake incapacitation for reform. A man in prison cannot burglarize your house. That is a real benefit, and it is the reason the question of how long to hold the repeat offender is worth asking. But incapacitation ends when the sentence does, and the BJS cohort shows what comes next. A system that relies on it alone is only postponing the arrest. For a look at a state that tried a different mix, see Texas Chose Treatment Beds Over New Prisons in 2007, and Here Is Why.

Count honestly. Politicians love a statistic the way a toddler loves a stick: any statistic, held at any angle, swung at whoever is nearest. Policymakers love to announce that a program "cut recidivism by 10 percent" without saying which of the three measures they used. Ask which one. Ask whether the comparison group was truly comparable. Ask whether the follow-up lasted one year or nine. The BJS study is valuable precisely because its follow-up is long enough to see the curve flatten.

The founders' view of punishment

The Founders were not sentimental about crime, and they were not cruel about it. The Eighth Amendment forbids "cruel and unusual punishments," which shows that they thought punishment should have a limit. The same generation built penitentiaries, a Quaker-influenced experiment in Pennsylvania, because they hoped that solitude and labor would produce penitence. The word "penitentiary" tells you what they wanted: a place for repentance, not only for confinement. Whether the experiment worked is a fair argument. That its purpose was reform as well as retribution is not in doubt.

So the question the BJS table puts to us is a very old one, and a very American one: how do you hold a man fully responsible for what he did and still leave a door open for what he might become? The data's answer is not that the door should be closed, and it is not that the door should be unguarded. It is that you should stand at the door in the first year, when most of the trouble happens, and that you should know which of the men walking through it are the ones to watch.

The verdict

Reformers should drop the claim that most released prisoners go straight, because the arrest data do not support it. Hard-liners should drop the claim that the numbers prove nobody changes, because the curve flattens, the age pattern is steep, and a large minority of releasees are never arrested again. Both camps should accept the thing the arithmetic of the repeat offender teaches: policy is won or lost on the few who generate most of the offenses, and on the first year after release.

That is a conservative conclusion in the best sense. It is skeptical of slogans, attentive to incentives, respectful of the victim and hopeful about the man. It asks for evidence before it asks for money. And it is what you get when you read the whole table instead of the line that flatters your side.

The case in brief

1. The Bureau of Justice Statistics followed state prisoners released in 2005 for nine years and found that 44 percent were arrested within a year, 68 percent within three and 83 percent within nine.
2. Arrest is a blunt measure: it is not a conviction, and it undercounts crime, so quote it with its limits.
3. A small group of repeat offenders accounts for a large share of all arrests, so the headline rate is the wrong target and the heavy offenders are the right one.
4. Risk is front-loaded, so supervision, housing and work belong in the first year, and parole should reward earned change and be explained in writing.
5. Redemption and accountability are partners, not rivals: release should be earned, supported and measured.

What you can do: ask your state corrections department or parole board for its published recidivism report and see which measure, arrest, conviction or return to prison, it uses and how long it follows releasees.