Corrections & Punishment

A number that means four different things

The Bureau of Justice Statistics followed 401,288 people released from state prisons in 30 states in 2005. Within three years, 68 percent had been arrested for a new offence. Within nine years, 83 percent had.

Now the question this lesson is really about. Does that number show that the prison system failed?

Your answer depends entirely on what you think the prison was for, and there are four serious answers to that, held by serious people, which disagree. Under one of them the 83 percent figure is a devastating indictment. Under another it is almost completely irrelevant to whether the sentences were justified. If you cannot say which is which, you cannot read a correctional policy argument, because most such arguments are two people using different justifications and neither one noticing.

Retribution: punishment as deserved

Retribution holds that punishment is justified because it is deserved. A person who culpably commits a wrong deserves to suffer a proportionate hardship in response, and the state is the agent that imposes it. That is the whole claim. Notice what it does not contain: any reference to future consequences.

Kant made the point with a thought experiment that still does the work. Imagine an island society dissolving itself, its people dispersing tomorrow. On Kant's view the last murderer in prison must still be executed before they go, because the punishment is owed regardless of any deterrent or protective effect, of which there can now be none.

Modern retributivism, developed by Andrew von Hirsch and others under the heading of just deserts, adds a systematic idea of proportionality: the severity of the punishment should track the seriousness of the offence, measured by harm done and culpability. Sentencing guidelines with offence-severity grids are, structurally, an attempt to operationalise this.

Two features of retribution are consistently missed by students and by politicians.

First, it is a limiting principle as well as a licensing one. If desert sets the amount, then punishing more than is deserved is unjust even when doing so would prevent crime. A retributivist has a principled objection to preventive detention of the dangerous-but-not-yet-guilty, and to a life sentence for shoplifting, that a pure consequentialist lacks. Some of the sharpest critics of long American sentences are retributivists.

Second, retribution makes no empirical prediction about crime. You cannot refute it with recidivism data. That is either its great virtue, since it does not depend on contested social science, or its great weakness, since it insulates itself from evidence. Both readings are defensible.

Key idea: Retribution is backward-looking. It asks what the offender deserves for what they did, sets both a floor and a ceiling on punishment, and generates no testable claim about future offending.

Deterrence: punishment as a price

Deterrence is forward-looking and consequentialist. Punishment is justified because the threat of it prevents offending. It has two branches that must be evaluated separately.

General deterrence operates on the public: people who are not punished refrain from offending because they observe what happens to those who are. Specific deterrence operates on the punished individual: having experienced the sanction, they are less inclined to repeat.

Beccaria set out the three levers in 1764, and they remain the framework: certainty, the probability of being caught and sanctioned; severity, how much the sanction costs; and celerity, how quickly it follows the act.

Here the theory earns its keep, because it makes sharp predictions that have been tested extensively. Daniel Nagin's reviews of that literature reach a consistent conclusion: certainty of apprehension has substantial deterrent effects, while increases in sentence severity produce small effects at best, and often none that can be detected. People who commit crimes are typically not performing accurate expected-value calculations over sentence lengths they do not know.

That finding has an uncomfortable implication for American policy, because nearly every major sentencing reform of the 1980s and 1990s pulled the severity lever, which the evidence says is the weak one, rather than the certainty lever, which is the strong one.

Incapacitation: punishment as physical prevention

Incapacitation makes the simplest claim of the four. A person in a cell cannot burgle a house on the outside. The crime prevented is not a matter of persuasion or reform; it is a matter of walls.

The effect is real and nobody disputes it exists. The entire argument is about magnitude, and the magnitude depends on three quantities.

The first is the individual offending rate of the person confined, which criminologists write as lambda. Confining someone who would have committed forty burglaries a year prevents forty burglaries. Confining someone who would have committed one prevents one.

The second is who the marginal prisoner is. A system that imprisons relatively few people imprisons the highest-rate offenders. As it expands, it necessarily reaches people with lower rates, so each additional prison bed buys less crime prevention than the last. This is the source of the diminishing returns finding that the National Research Council's 2014 consensus report reached about American incarceration at its current scale.

The third is replacement. In organised illegal markets, removing a participant may simply create a vacancy that someone else fills, so the market's total offending is unchanged. This applies most strongly to drug distribution and least to impulsive interpersonal violence.

Incapacitation also runs directly into the age-crime curve. Offending peaks in the late teens and early twenties and declines steeply thereafter. A sentence long enough to hold someone into their fifties is incapacitating a person whose expected offending rate has already fallen close to zero, which is why very long sentences are hard to defend on incapacitation grounds even when they are easy to defend on retributive ones.

Rehabilitation: punishment as an occasion for change

Rehabilitation holds that the sanction is justified by the change it produces in the person, so that they offend less after it than they would have. Unlike deterrence it does not work through fear; it works through treating a criminogenic need, whether that is an untreated substance use disorder, illiteracy, antisocial thinking patterns, or the absence of any employable skill.

This is the most directly testable of the four, because the prediction is precisely a lower post-release offending rate than the counterfactual, and the counterfactual can sometimes be constructed by randomisation.

It also has the most dramatic history, which the next lesson covers: an era of near-universal adoption, a collapse in the 1970s produced by a single influential review, and a partial recovery driven by better evaluation methods that found real effects the earlier review had missed.

The comparison table, which is the point of the lesson

Justification	Direction	What justifies the sanction	Empirical prediction	What would count against it
Retribution	Backward	The offender deserves it, proportionate to harm and culpability	None about crime rates	Nothing empirical; only an argument that desert does not justify state hardship
Deterrence	Forward	The threat prevents offending by others and by the offender	Offending falls as certainty, severity, or celerity rises	Severity increases that produce no measurable change in offending
Incapacitation	Forward	Confinement physically prevents offences the person would have committed	Crime falls in proportion to the offending rate of those confined	Expansion that prevents little crime, or replacement in illegal markets
Rehabilitation	Forward	The person is changed so they offend less afterward	Lower post-release offending than the counterfactual	Evaluations showing no reduction relative to a comparable control group

What matters here: Three of the four justifications generate predictions you can test with data, and one does not. That is not a defect in retribution; it is a statement about what kind of claim it makes. Confusing the two kinds is the most common error in correctional policy argument.

One case, four answers

Work a concrete example, because the abstractions dissolve on contact with a real sentence.

A man is 65 years old. He is serving year 30 of a life sentence imposed for a murder he committed at 22. He has no disciplinary infractions in eighteen years. He has completed every available programme. A parole board is deciding whether to release him.

Retribution asks whether thirty years is proportionate to the harm and culpability of the original offence, and holds that the answer does not change because he has behaved well. Good conduct is not restitution for a death. A retributivist could support release if the desert has been discharged, or oppose it if it has not, but on neither branch does his current dangerousness matter.

General deterrence asks whether releasing him weakens the threat facing others. The honest answer is almost certainly not, since parole decisions thirty years after the fact are not visible to potential offenders and severity effects are weak in any case.

Incapacitation asks his expected offending rate at 65. The age-crime curve says it is very low. On this justification alone he should have been released some years ago.

Rehabilitation asks whether he has changed. The record suggests he has, and rehabilitation would treat his release as the successful conclusion of the sanction.

Three of the four point toward release. One is indifferent to everything that has happened since 1982. A real parole board contains people using all four without labelling them, which is why parole hearings so often feel like conversations at cross purposes.

A fifth position, and the mixed view

Restorative justice reframes the question. It treats crime primarily as harm to people and relationships rather than as an offence against the state, and asks what would repair it. In practice this means facilitated conferences involving the victim, the offender, and affected community members, producing agreements about restitution and amends. Reviews by Lawrence Sherman and Heather Strang of randomised and quasi-experimental studies find modest reductions in repeat offending and, more consistently, substantially higher victim satisfaction than conventional processing produces. That second finding matters, because conventional criminal justice serves victims badly and rarely measures whether it did.

Finally, almost no serious theorist holds one justification purely. H. L. A. Hart's influential move was to separate two questions that had been run together. Why have a practice of punishment at all? Answer: because it reduces crime, a consequentialist answer. Whom may we punish, and how much? Answer: only the guilty, and only in proportion to desert, a retributive answer. On this mixed view, consequences justify the institution and desert constrains its distribution, which is roughly how modern sentencing systems are actually built.

Common misconceptions

• Retribution just means revenge. Revenge is unlimited and personal; retribution is proportionate, impersonal, and imposes a ceiling on punishment as well as a floor.
• A high recidivism rate proves punishment does not work. It bears directly on rehabilitation and specific deterrence, indirectly on incapacitation, and not at all on retribution.
• Longer sentences deter more. Reviews consistently find that certainty of apprehension deters and that severity increases produce small or undetectable effects.
• Incapacitation always prevents crime efficiently. Its yield depends on the offending rate of the marginal prisoner, and falls sharply as a system expands and as prisoners age.
• You must choose one justification. Hart's mixed view, in which consequences justify the practice and desert limits its distribution, is closer to how sentencing systems are actually constructed.

Putting it together

• Of the 401,288 people released from state prisons in 30 states in 2005, 68 percent were arrested within three years and 83 percent within nine.
• Retribution is backward-looking, sets both a floor and a ceiling on punishment, and makes no prediction about crime rates.
• Deterrence predicts that offending falls with certainty, severity, and celerity, and the evidence supports certainty far more strongly than severity.
• Incapacitation is real but its yield depends on the offending rate of the marginal prisoner, on replacement in illegal markets, and on the steep decline of offending with age.
• Rehabilitation makes the most directly testable claim, that post-release offending is lower than the counterfactual, which randomised and quasi-experimental designs can assess.
• Hart's mixed view assigns the general justifying aim to consequences and the distribution of punishment to desert, which describes most modern sentencing systems.

Sources

1. Bureau of Justice Statistics. (n.d.). Corrections statistics and recidivism research. Office of Justice Programs. bjs.ojp.gov
2. National Research Council. (2014). The growth of incarceration in the United States: Exploring causes and consequences. Washington, DC: National Academies Press. nap.nationalacademies.org
3. Nagin, D. S. (2013). Deterrence in the twenty-first century. Crime and Justice, 42(1), 199-263. doi.org
4. Stanford Encyclopedia of Philosophy. (n.d.). Legal punishment. Stanford University. plato.stanford.edu
5. Hart, H. L. A. (1968). Punishment and responsibility: Essays in the philosophy of law. Oxford: Oxford University Press.
6. von Hirsch, A. (1976). Doing justice: The choice of punishments. New York: Hill and Wang.

Prisoner number one

In October 1829, Charles Williams, an eighteen-year-old farmer convicted of burglary, walked into Eastern State Penitentiary in Philadelphia as its first inmate. He was given a cell with a private exercise yard, a hooded escort so that he would see no other prisoner's face, a Bible, and work he could do alone. He was to serve two years in near-total isolation, so that, in the theory of the men who built the place, he would look inward, confront what he had done, and repent.

The word penitentiary comes from penitence. That is not decoration; it is the design specification.

What makes the building historically important is not its architecture, striking as John Haviland's radial plan was, with cell blocks radiating from a central rotunda so that a single guard could survey every corridor. It is that imprisonment as the standard punishment for serious crime was, at that moment, a new idea.

What the penitentiary replaced

For most of European and colonial American history, a jail was a holding pen. It held people awaiting trial, people awaiting execution, and debtors. It was not, in itself, the punishment.

The punishments were public and physical. The stocks and the pillory. Whipping. Branding. Fines. Transportation to a colony. And, for a very long list of offences, hanging: the English criminal law of the eighteenth century, later called the Bloody Code, prescribed death for more than two hundred offences, many of them property crimes.

The Enlightenment critique of this system came in three influential books within thirty years. Cesare Beccaria's On Crimes and Punishments, published in 1764, argued that punishment should be proportionate, certain, and prompt rather than savage, and that torture and the death penalty were both cruel and ineffective. John Howard, high sheriff of Bedfordshire, toured English and continental prisons and published The State of the Prisons in 1777, documenting filth, disease, and the practice of charging prisoners fees for their own confinement. Jeremy Bentham designed the Panopticon in 1791, a circular building in which inmates could always be observed and never know when they were being watched.

The American institutional response began at the Walnut Street Jail in Philadelphia, where in 1790 a penitentiary house was established under Quaker influence: solitary cells, labour, scripture, silence. Eastern State was the mature version of the same theory.

Key idea: Imprisonment as punishment is roughly two hundred years old. Before the penitentiary, confinement was mostly a waiting room for corporal punishment, execution, or transportation, and the reformers who invented the prison thought they were replacing barbarism with a humane and rational alternative.

Two systems, one argument

Two rival designs competed in the early nineteenth century, and the argument between them is more interesting than the outcome.

The Pennsylvania system, also called the separate system, kept prisoners in individual cells at all times, working, eating, and worshipping alone. The reasoning was that contact between prisoners spreads criminal habits, and that solitude combined with labour and religious instruction produces genuine reflection.

The Auburn system, developed in New York in the 1820s, kept prisoners isolated at night but had them work together by day under a rule of absolute silence, enforced by the lash. Movement between areas used the lockstep, a shuffling formation with each man's hand on the shoulder ahead.

	Pennsylvania, separate	Auburn, congregate
By day	Alone in cell, working alone	Group workshops under enforced silence
Rationale	Contact contaminates; solitude produces reflection	Contact is manageable if speech is prevented; labour disciplines
Economics	Expensive: large cells, individual yards, low output	Cheaper: small cells, factory-scale production, revenue from contracts
Outcome	Widely admired in Europe, largely abandoned in America	Became the American model

Auburn won in the United States for a reason that recurs throughout this course: it was cheaper, and it produced revenue. Prison labour under contract to private manufacturers made congregate institutions financially defensible in a way that solitary cells with individual yards never were.

The separate system also drew early and serious criticism on its effects. Charles Dickens visited Eastern State in 1842 and left convinced that prolonged isolation inflicted a slow and profound damage on the mind, worse in his judgement than physical punishment because it was invisible. That objection, made in 1842, is essentially the objection made in the modern literature on solitary confinement, which is the subject of a later lesson in this course.

The southern exception

The northern penitentiary story leaves out what happened in the South after 1865, and it cannot be left out.

The Thirteenth Amendment abolished slavery and involuntary servitude except as a punishment for crime whereof the party shall have been duly convicted. Southern states enacted Black Codes criminalising vagrancy, breaking a labour contract, and similar offences, and then leased the resulting convicts to planters, mine operators, and railroads. Under convict leasing the state received payment and the lessee received labour, with no incentive to keep workers alive, since a dead leased convict could be replaced at no cost to the lessee. Death rates in some leasing operations were catastrophic.

Other states ran plantation prisons directly, most notoriously Mississippi's Parchman Farm, where incarcerated men worked cotton under armed trusty guards drawn from the prison population itself. Douglas Blackmon's documentary history of this period is the standard account, and it establishes that for a large part of the American South, the institution that followed slavery was not the penitentiary of Philadelphia but something much closer to what preceded it.

The rehabilitative ideal

The next transformation was intellectual. In 1876, Zebulon Brockway opened the Elmira Reformatory in New York on a new premise: that the sentence should be indeterminate, its end depending on the prisoner's progress rather than on the offence alone. Elmira used a marks system, education, trade training, and release on parole under supervision.

Over the following decades this became the American consensus, extended by the Progressive era institutions of probation, the juvenile court founded in Cook County, Illinois in 1899, and parole boards in nearly every state. The underlying model was medical: crime is a symptom, the offender is diagnosed, treatment is prescribed, and release follows cure. Indeterminate sentences of one to fifteen years or five to life were normal, with the actual release date decided by a parole board reading a file.

By the 1950s and 1960s this was so dominant that it barely needed defending. Then it collapsed in about a decade.

What Martinson actually said

In 1974 Robert Martinson published an article in The Public Interest summarising a large survey he had conducted with Douglas Lipton and Judith Wilks of 231 evaluations of correctional treatment programmes published between 1945 and 1967. His summary sentence was that with few and isolated exceptions, the rehabilitative efforts reported so far had had no appreciable effect on recidivism.

Two things happened to that sentence. It was compressed in public discussion into the slogan nothing works, which is stronger than what he wrote. And Martinson himself published a substantial revision in 1979 in the Hofstra Law Review, concluding on further analysis that some treatments were beneficial and that his earlier conclusion had been too sweeping. The retraction received a small fraction of the attention the original had.

Why did the original land so hard? Because it arrived at a moment when critics on both sides of the political spectrum were already looking for reasons to abandon the rehabilitative model, and it gave each of them one.

What matters here: Martinson's 1974 review was more qualified than the nothing works slogan built from it, and he revised his conclusion in 1979. Both facts are routinely omitted, and the omission has shaped correctional policy for fifty years.

The convergence that produced determinate sentencing

The attack on indeterminate sentencing came from directions that rarely agree.

From the left, the objection was arbitrariness and coercion. The American Friends Service Committee's 1971 report Struggle for Justice argued that indeterminate sentencing gave enormous unreviewable power to parole boards, that release depended on performing rehabilitation rather than on justice, and that the resulting disparities fell hardest on the poor and on Black prisoners. Judge Marvin Frankel's 1973 book Criminal Sentences: Law Without Order made the legal version of the argument: two judges could impose wildly different sentences on identical cases, with no standards and no appellate review, which he described as lawlessness at the heart of the system.

The uprising at Attica in September 1971 sharpened everything. Some 1,281 prisoners took control of the New York facility and held hostages for four days. When state police retook it by force, 43 people died, including ten hostages killed by state gunfire. The event made the internal conditions of American prisons a matter of national politics in a way they had not been.

From the right, the objection was leniency: parole boards released people too early, sentences did not mean what they said, and the public could not know what a stated sentence actually was.

Both critiques pointed at the same remedy. Maine abolished parole release in 1976. California enacted a Uniform Determinate Sentencing Act the same year, replacing the indeterminate model with fixed terms. Minnesota established a sentencing guidelines commission in 1980. Congress passed the Sentencing Reform Act in 1984, creating the United States Sentencing Commission and abolishing federal parole.

What followed was not what the reformers on the left had expected. Determinate sentencing removed the release valve while legislatures raised the terms, and the American prison population, which had held near 200,000 for decades, began the climb that took it past 1.6 million by 2009.

The partial recovery

Meanwhile the research the nothing works slogan had buried was being redone with better methods. Meta-analysis, which pools effect sizes across studies rather than counting significant results, was applied to correctional treatment through the 1980s and 1990s by Mark Lipsey, Don Andrews, James Bonta, and others. The picture that emerged was not that everything works, which would have been equally useless. It was that specific programmes with specific properties produce measurable reductions in reoffending, while others produce nothing and a few make things worse. That literature, and the risk-need-responsivity model built from it, is the subject of a later lesson.

Common misconceptions

• Prisons have always been the main punishment for crime. Imprisonment as a sentence is roughly two centuries old; before it, jails mostly held people awaiting trial, execution, or transportation.
• The penitentiary was invented by people indifferent to prisoners. It was invented by reformers who considered it a humane replacement for the gallows and the whipping post, which is what makes its later history instructive.
• The Auburn system won because it worked better. It won because it was cheaper and produced contract labour revenue that the separate system could not.
• Martinson proved that rehabilitation does not work. His 1974 conclusion was more qualified than the slogan, and he substantially revised it in 1979.
• Determinate sentencing was a conservative project. Prison reformers and civil liberties organisations attacked indeterminate sentencing first, on arbitrariness grounds, and legislatures then set the fixed terms high.

Looking back

• Eastern State Penitentiary received its first prisoner in October 1829, built on the theory that solitude, labour, and scripture produce penitence.
• Before the penitentiary, punishment was corporal, capital, or transportation, and Beccaria, Howard, and Bentham supplied the reform argument that produced the prison.
• The Pennsylvania separate system and the Auburn congregate system competed on theory, and Auburn prevailed on cost and contract labour revenue.
• The Thirteenth Amendment's punishment exception enabled convict leasing and plantation prisons in the postbellum South, a distinct institutional history from the northern penitentiary.
• Elmira in 1876 introduced the indeterminate sentence and parole, and the rehabilitative ideal dominated American corrections until the 1970s.
• Martinson's 1974 review, his 1979 revision, the Attica uprising, and critiques from both left and right produced the determinate sentencing turn, after which the prison population grew from about 200,000 to more than 1.6 million.

Sources

1. Eastern State Penitentiary Historic Site. (n.d.). History of Eastern State Penitentiary. easternstate.org
2. Britannica. (2025). Prison. In Encyclopaedia Britannica. britannica.com
3. Bureau of Justice Statistics. (n.d.). Corrections. Office of Justice Programs. bjs.ojp.gov
4. Library of Congress. (n.d.). Digital collections on American prisons and reform. loc.gov
5. Martinson, R. (1974). What works? Questions and answers about prison reform. The Public Interest, 35, 22-54.
6. Blackmon, D. A. (2008). Slavery by another name: The re-enslavement of Black Americans from the Civil War to World War II. New York: Doubleday.

Two hundred and fifty-eight boxes

The federal sentencing table is a grid. Down the left side run 43 offence levels. Across the top run six criminal history categories. That produces 258 cells, and each cell contains a range in months. Offence level 26 in criminal history category I gives 63 to 78 months. Move one column right, to category II, and the same offence level gives 70 to 87.

A judge who wants to know what a defendant is facing calculates the offence level, calculates the criminal history category, reads the cell, and has a range with a top roughly 25 percent above its bottom. That is what replaced the old system, in which the judge had a statutory maximum, a great deal of unreviewable discretion, and a parole board that would decide the real release date years later.

This lesson is about how the grid got built, what the Supreme Court did to it, and the thing that most surprises students: the reform designed to constrain discretion did not eliminate it. It moved it.

Four ways to structure a sentence

Before the mechanics, the taxonomy, because states differ and the vocabulary is used loosely.

Structure	Who sets the actual time served	Example
Indeterminate	Judge sets a range; a parole board decides release within it	Most American states before the mid-1970s; many still
Determinate	Judge sets a fixed term; release follows automatically less good-time credit	California after 1976; the federal system after 1987
Presumptive guidelines	A commission sets ranges the judge must use absent stated reasons subject to appeal	Minnesota from 1980; the federal system from 1987 to 2005
Mandatory minimum	The legislature, by statute, regardless of guidelines or judge	Federal drug and firearm statutes; many state statutes

These stack. A federal drug defendant can face a guideline range, a statutory mandatory minimum that overrides the bottom of that range, and a consecutive firearm count, all at once.

How the federal grid is calculated

Work it through, because the mechanics matter more than the concept.

Start with a base offence level set by the type of crime. Add or subtract specific offence characteristics: drug quantity, loss amount, whether a weapon was possessed, the number of victims, the role in the offence. Then apply adjustments, including a reduction, usually two levels and sometimes three, for acceptance of responsibility, which in practice means pleading guilty.

Criminal history is scored separately by assigning points to prior sentences, more points for longer prior terms, with additional points for offences committed while under supervision. The total maps onto one of six categories.

One feature deserves its own paragraph because it startles people who assume the sentence tracks the conviction. Under the guidelines' relevant conduct provision, a judge may consider conduct that was never charged, and conduct on counts where the jury acquitted, provided it is proved to the judge by a preponderance of the evidence. A defendant convicted on one count and acquitted on three can have the acquitted conduct raise the guideline range on the count of conviction. Courts have upheld this repeatedly, and the Sentencing Commission has narrowed but not abolished it.

Key idea: The guideline range is produced by a calculation over quantities and characteristics, many of which the jury never decided, using a standard of proof far below the one that governed the trial.

The constitutional line: Apprendi to Booker

That last feature is exactly what the Supreme Court spent a decade addressing.

In Apprendi v. New Jersey, decided in 2000, the Court held that any fact other than a prior conviction that increases the penalty for a crime beyond the prescribed statutory maximum must be submitted to a jury and proved beyond a reasonable doubt. Charles Apprendi had pleaded guilty to a firearms count carrying a maximum of ten years, and the judge had then found by a preponderance that the shooting was racially motivated and imposed twelve under a hate crime enhancement. The Court reversed.

The obvious question was what counted as the statutory maximum. In Blakely v. Washington, decided in 2004, the Court answered: the relevant maximum is the maximum sentence a judge may impose on the facts found by the jury or admitted by the defendant, not the higher number printed in the statute. Ralph Blakely had pleaded to a kidnapping count with a standard range topping out at 53 months, and the judge had imposed 90 after finding deliberate cruelty. That, the Court held, was unconstitutional.

Blakely concerned Washington State, and everyone understood immediately what it implied for the federal guidelines, which worked the same way. The Court took the question up within months.

United States v. Booker, decided in January 2005, produced two majorities with different memberships. One held that the mandatory federal guidelines violated the Sixth Amendment for the reasons given in Blakely. The other chose the remedy: rather than requiring juries to find guideline facts, it severed the statutory provision that made the guidelines binding. The guidelines became advisory. Judges must still calculate the range and consider it, and appellate courts review sentences for reasonableness, but a judge may impose a sentence outside the range and explain why.

Note the shape of that remedy. The constitutional problem was that judges were finding facts that raised sentences. The solution was not to stop them finding facts; it was to stop the findings from binding them. That is a real fix to the Sixth Amendment problem and it does nothing about the preponderance standard or relevant conduct.

Mandatory minimums, and who really controls them

Guidelines are ranges. Mandatory minimums are floors, and they sit above the guidelines when they conflict.

The Anti-Drug Abuse Act of 1986 established the federal drug minimums that still dominate the system, keyed to drug quantity: five years for one set of thresholds, ten for another. It also created the notorious ratio between crack and powder cocaine. Five grams of crack triggered the same five-year minimum as 500 grams of powder, a 100 to 1 quantity ratio for two forms of the same drug. The Fair Sentencing Act of 2010 reduced that ratio to about 18 to 1 and eliminated the mandatory minimum for simple possession of crack; the First Step Act of 2018 made the 2010 changes retroactive.

Here is the part that reorganises how you should think about sentencing reform. A mandatory minimum is triggered by what the prosecutor charges. Charge a quantity above the threshold and the floor applies; charge below it and it does not. The same is true of the enhancements that stack additional mandatory terms. So a statute that removes discretion from judges hands it to prosecutors, who exercise it without a hearing, without a record, and without appellate review.

The plea system completes the transfer. Around 97 percent of federal convictions and roughly 94 to 95 percent of state felony convictions are obtained by guilty plea rather than trial. When the difference between the plea offer and the exposure at trial is measured in decades, the charging decision is the sentencing decision, and it is made in an office rather than a courtroom.

The upshot: Determinate sentencing and mandatory minimums did not abolish discretion. They relocated it from judges and parole boards, whose decisions were made on a record and could be reviewed, to prosecutors, whose charging decisions are largely unreviewable.

Three strikes and truth in sentencing

Two other structural changes belong here.

Habitual offender laws, popularly called three strikes, escalate the sentence for a new offence based on prior convictions. Washington enacted the first modern version in 1993 and California followed in 1994 with by far the broadest, allowing any felony to serve as the triggering third strike. In Ewing v. California, decided in 2003, the Supreme Court rejected an Eighth Amendment challenge to a sentence of 25 years to life imposed for shoplifting golf clubs by a defendant with prior serious convictions. California voters narrowed the law by initiative in 2012 to require that the third strike itself be serious or violent.

Truth in sentencing addressed the gap between the announced sentence and time actually served. The federal crime bill of 1994 offered states grant funding conditioned on requiring people convicted of violent offences to serve at least 85 percent of the imposed term. Many states adopted the standard. The reform did what it said: sentences became more transparent. It also lengthened time served substantially, because the announced numbers had been set in an era when everyone understood them to be discounted.

Did the guidelines reduce disparity?

Partly, and the honest answer requires distinguishing kinds of disparity.

Guidelines clearly reduced inter-judge disparity in the sense of two judges in the same courthouse imposing very different sentences on similar cases, which was Judge Frankel's original complaint. Research after Booker found that inter-judge variation increased once the guidelines became advisory, which is evidence that they had been doing some of the work claimed for them.

They did much less about demographic disparity, and the Sentencing Commission's own periodic reports have documented it. Its analyses have found that Black male defendants receive sentences on the order of a fifth longer than similarly situated white male defendants, with the gap concentrated in the decisions the guidelines do not control: the departure and variance decisions and, upstream, charging. Separate research by Sonja Starr and M. Marit Rehavi, tracing federal cases from arrest onward rather than from sentencing, found that a large share of the final disparity is generated by charging decisions before the guidelines are ever consulted.

That is exactly what the relocation-of-discretion argument predicts. A system that constrains the visible decision and leaves the upstream one unconstrained will produce disparity in the upstream one.

Common misconceptions

• The guidelines tell the judge exactly what sentence to impose. They produce a range, they have been advisory in the federal system since 2005, and the judge may vary from them with reasons.
• A defendant is sentenced only for what the jury convicted them of. Relevant conduct allows uncharged and even acquitted conduct to raise the guideline range if proved to the judge by a preponderance.
• Booker struck down the sentencing guidelines. It held the mandatory application unconstitutional and remedied that by severing the binding provision, leaving the guidelines in force as advisory.
• Mandatory minimums remove discretion from the system. They remove it from judges and give it to prosecutors, whose charging decisions determine whether a floor applies at all.
• Truth in sentencing did not change how long people serve. Requiring 85 percent of a term that had been set assuming substantial discounting lengthened actual time served considerably.

What to carry forward

• The federal sentencing table has 43 offence levels and six criminal history categories, and each of its 258 cells contains a range whose top is roughly a quarter above its bottom.
• Relevant conduct lets judges consider uncharged and acquitted conduct at a preponderance standard, which is the practice the Sixth Amendment cases circled without abolishing.
• Apprendi required jury findings for facts raising a sentence above the statutory maximum; Blakely defined that maximum as what the jury findings alone permit; Booker applied this federally and made the guidelines advisory.
• Mandatory minimums override guideline ranges from below and are triggered by charging decisions, which transfers discretion from judges to prosecutors.
• Roughly 97 percent of federal convictions come by guilty plea, so the charging decision is in practice the sentencing decision.
• Guidelines reduced inter-judge disparity, and Sentencing Commission and independent research find continuing demographic disparity concentrated in charging, departures, and variances.

Sources

1. United States Sentencing Commission. (n.d.). Guidelines manual and sentencing table. ussc.gov
2. Apprendi v. New Jersey, 530 U.S. 466 (2000). supreme.justia.com
3. Blakely v. Washington, 542 U.S. 296 (2004). supreme.justia.com
4. United States v. Booker, 543 U.S. 220 (2005). supreme.justia.com
5. Ewing v. California, 538 U.S. 11 (2003). supreme.justia.com
6. Frankel, M. E. (1973). Criminal sentences: Law without order. New York: Hill and Wang.

Seven in ten are not serving a sentence

At midyear 2022 local jails in the United States held roughly 660,000 people. Around seven in ten of them had not been convicted of the offence for which they were being held. They were legally innocent, awaiting the resolution of a case, and in jail because they could not post the amount of money a court had set.

That single fact separates jails from prisons more sharply than anything about buildings or uniforms, and it is the reason this lesson treats them as two institutions rather than two sizes of the same one.

The correctional population, laid out

Take the whole system first, so the parts have somewhere to sit. At the end of 2022, roughly 5.5 million adults in the United States were under some form of correctional control.

Component	Approximate number	Run by	Who is held or supervised
State and federal prisons	About 1.2 million	State departments of corrections and the federal Bureau of Prisons	People sentenced to more than a year
Local jails	About 660,000	Counties and cities, usually the elected sheriff	Mostly pretrial detainees, plus short sentences and holds
Probation and parole	About 3.7 million	State and county agencies, federal probation	People supervised in the community

Two observations follow immediately. Community supervision holds more than twice as many people as all prisons and jails combined, which is why a course that spends all its attention on cells is describing a minority of the system. And local jails, run by thousands of separate county authorities with no common standards, hold a population that most people conflate with prisons.

The differences that matter

	Jail	Prison
Authority	County or municipal, usually an elected sheriff	State or federal executive agency
Legal status of those held	Mostly unconvicted, awaiting case resolution	Convicted and sentenced
Typical stay	Days to weeks; the median is very short	Years
Annual admissions	Millions, with very high turnover	Several hundred thousand
Programming	Minimal; stays are too short and unpredictable to plan around	Education, work, and treatment programmes of varying quality
Classification	Limited; mixed population in one facility	Security levels, from minimum camps to maximum custody
Leading cause of death	Suicide accounts for a large share	Illness, reflecting an older and sicker population

The turnover row is the one to sit with. Jails admit people millions of times a year, an order of magnitude more often than prisons do. Most of those stays are brief. But brief is not harmless: three days in jail is long enough to lose a job, miss a rent payment, fail to collect a child from school, or lose housing. The system's most common experience of incarceration is a short jail stay, and it is almost invisible in datasets built around prison sentences.

Key idea: Jails are high-turnover local institutions holding mostly unconvicted people for short periods. Prisons are lower-turnover state institutions holding sentenced people for years. Statements about incarceration that do not specify which are usually wrong about one of them.

Why people are in jail: money

After an arrest, a court decides whether to release the person pending trial and on what conditions. In the traditional American system that decision is usually expressed as a money bail amount. Post it and you go home; fail to post it and you stay.

The consequence is that pretrial detention sorts on ability to pay rather than on risk. Two people charged with the same offence, assessed as equally likely to appear, are separated by whether one of them can raise a few hundred dollars.

For a long time the defence of this arrangement was that detention merely holds people who would otherwise flee, and does not affect the outcome of their cases. That claim is testable, and it has been tested.

The bail judge design

The obstacle to studying pretrial detention is obvious: judges detain the people they consider riskiest, so detained defendants differ from released ones in exactly the ways that predict bad outcomes. Comparing them directly measures the judge's judgement, not the effect of detention.

The solution exploits how courts assign cases. In many jurisdictions, arrestees are assigned to whichever bail magistrate is on duty, effectively at random, and magistrates differ substantially in how readily they release. That difference in leniency is a randomly assigned nudge toward release, uncorrelated with the defendant's own characteristics, which is precisely what an instrumental variable requires.

Will Dobbie, Jacob Goldin, and Crystal Yang applied this design to data from Philadelphia and Miami. They found that pretrial release substantially reduced the probability of conviction, largely by reducing guilty pleas, and increased later employment. Related work by Paul Heaton, Sandra Mayson, and Megan Stevenson on misdemeanour cases in Harris County, Texas found that detained defendants were considerably more likely to be convicted, more likely to receive a jail sentence, and more likely to be charged with new offences afterward.

The mechanism is not mysterious. A detained defendant has less bargaining power, because a plea that ends the case today ends the detention today, and because time served often equals or exceeds the offered sentence. A detained defendant also loses employment and housing, which are the things that most reliably predict not reoffending.

What matters here: Pretrial detention is not a neutral holding operation. Using random assignment to bail magistrates as an instrument, researchers find it causally increases conviction, increases guilty pleas, and reduces later employment.

New Jersey provides the policy-scale counterpart. The state largely eliminated money bail in 2017, replacing it with a risk-based assessment and, for the highest-risk cases, a motion for detention without any monetary option. The pretrial jail population fell substantially, and monitoring found that appearance rates and rates of new criminal activity while on release were broadly unchanged. That is one state with one design, and it is the single most useful natural experiment available on the question.

Who is in prison, and why the state and federal answers differ

The composition of prison populations is where most public argument goes wrong, because people generalise from the federal system, which is the atypical one.

The federal Bureau of Prisons holds roughly 155,000 to 160,000 people, which is about an eighth of the national total. Its population is dominated by drug offences, which account for something close to half of it, along with weapons and immigration offences, because those are the areas of federal jurisdiction.

State prisons hold the other seven eighths. There, drug offences account for roughly one in seven people, and offences classified as violent account for more than half.

The implication is arithmetic and it is uncomfortable for a popular reform slogan. Releasing everyone imprisoned for a drug offence in the United States would reduce the prison population by well under a quarter. A policy that does not touch sentences for violent offences cannot substantially reduce American incarceration, because that is where most of the people are. This is not an argument against drug sentencing reform, which has its own justifications. It is an argument about arithmetic that any serious decarceration proposal has to confront, and the final lesson of this course returns to it.

Length of stay, age, and who is actually there

A few further compositional facts change how the population looks.

Time served is highly skewed. Many people leave state prison within a year or two, while a long-serving minority accumulates in the population, so the people in prison on any given day are serving much longer sentences on average than the people admitted in a given year. That difference between stock and flow explains why admissions can fall while the population does not.

The population is ageing, and rapidly. The share of state prisoners aged 55 and over has risen several-fold since the early 1990s, a direct consequence of long sentences imposed decades ago. Older prisoners cost far more to hold, because of medical care, and have the lowest reoffending rates of any group, which sets up a policy tension the parole lesson returns to.

Women are a small share of both populations, around one in fourteen prisoners and roughly one in ten of the jail population, but the female jail population grew faster than the male one over recent decades, and the great majority of women in jail are unconvicted.

Common misconceptions

• Jail and prison are interchangeable words. They are different institutions with different authorities, populations, legal postures, and lengths of stay.
• People in jail have been convicted. Roughly seven in ten have not; they are detained pending case resolution, most often because they cannot post money bail.
• Pretrial detention has no effect on case outcomes. Randomised bail magistrate designs show it increases conviction and guilty pleas and reduces later employment.
• Most prisoners are serving time for drug offences. That is close to true federally and false in state prisons, which hold seven eighths of the total and where violent offences account for more than half.
• Short jail stays are minor. A few days is long enough to cost a job, a tenancy, or custody arrangements, and the short stay is the most common experience of incarceration in America.

The short version

• Roughly 5.5 million adults are under correctional control: about 1.2 million in prisons, about 660,000 in jails, and about 3.7 million on probation or parole.
• Around seven in ten jail detainees are unconvicted, and money bail sorts pretrial release on ability to pay rather than on assessed risk.
• Jails are locally run with very high turnover and minimal programming; prisons are state or federal, hold sentenced people for years, and classify by security level.
• Randomly assigned bail magistrate leniency shows pretrial detention causally raises conviction and guilty plea rates and lowers later employment.
• New Jersey's 2017 elimination of money bail cut the pretrial jail population with broadly unchanged appearance and rearrest rates.
• Drug offences dominate the federal population but account for roughly one in seven state prisoners, while violent offences account for more than half.

Sources

1. Bureau of Justice Statistics. (n.d.). Corrections: Prisons, jails, probation, and parole. Office of Justice Programs. bjs.ojp.gov
2. Federal Bureau of Prisons. (n.d.). Statistics. U.S. Department of Justice. bop.gov
3. Dobbie, W., Goldin, J., and Yang, C. S. (2018). The effects of pretrial detention on conviction, future crime, and employment: Evidence from randomly assigned judges. American Economic Review, 108(2), 201-240. doi.org
4. Vera Institute of Justice. (n.d.). Research on jails and pretrial justice. vera.org
5. Prison Policy Initiative. (n.d.). Data on the whole correctional system. prisonpolicy.org

Five deprivations, catalogued from the inside

Gresham Sykes spent three years observing the New Jersey State Prison at Trenton, a maximum security institution, and published The Society of Captives in 1958. Its most durable contribution is a list. Sykes argued that imprisonment inflicts five distinct deprivations, and that almost everything about prison social life is an adaptation to them.

The deprivation of liberty is the obvious one, but Sykes emphasised the part that is not obvious: it is not only confinement to a place, it is confinement away from everyone, and it carries the moral rejection of the wider society, which is felt continuously.

The deprivation of goods and services reduces a person to a standard-issue material existence in a society that measures worth by possessions. What matters is less the level than the fact that it is imposed and equalised.

The deprivation of intimate relationships removes not only sexual life but the ordinary daily contact through which people confirm who they are.

The deprivation of autonomy is the one prisoners in Sykes's account resented most bitterly. It is not simply having rules; it is having rules whose reasons are not explained, imposed by officials who need not justify them, over matters as small as when to stand and where to sit.

The deprivation of security is the least expected. A person confined among others convicted of violence, unable to leave, unable to choose company, lives with the constant question of whether they will be tested and how they will respond.

Key idea: Sykes's five pains are not a list of hardships. They are a set of problems that the prison population must solve, and the culture, hierarchy, and economy that develop inside are the solutions.

The defects of total power

Sykes's second major observation is easy to miss and more important than the list.

On paper the prison is a system of total authority: officers command, prisoners obey, and the officers hold every formal power. In practice, Sykes found, officers cannot run a cell block that way. There are far more prisoners than officers, the officers are unarmed inside the block, and the daily job requires cooperation, moving people to work, meals, and counts, on time, without incident.

Officers therefore trade. They overlook minor rule violations in exchange for order. They allow informal privileges to prisoners who keep their own tier quiet. They avoid writing reports that would generate paperwork and antagonism. Sykes called this the corruption of authority, and it means that the actual operating rules of a prison are negotiated rather than imposed.

Every subsequent finding in this lesson depends on that insight. When you read that a prison has a policy, ask what the officers on the tier are actually trading for compliance.

Where the culture comes from

Sykes's explanation of inmate culture is a deprivation model: the pains of imprisonment are the cause, so any prison anywhere should generate broadly similar adaptations. The inmate code he described, do your own time, do not inform, do not exploit other prisoners, be tough, do not trust officers, is on this account a functional response to those five problems.

John Irwin and Donald Cressey challenged this in 1962 with an importation model. Prisoners, they pointed out, do not arrive as blank slates. They bring subcultures from the street. Irwin and Cressey described distinct orientations, including a thief subculture with its own long-standing norms about informing, a convict subculture oriented to doing time and gaining influence within the institution, and a straight orientation of people who see themselves as temporarily displaced from conventional life. On this account prison culture is largely imported, and it differs between institutions because their populations differ.

The dispute is not settled by choosing a side; it is settled by noticing that both mechanisms are real and that their relative weight varies. Institutions holding people from a single city with dense street networks look importation-heavy. Institutions with very restrictive regimes and mixed populations look deprivation-heavy. The useful question about any prison is which of the two is doing more work there, and why.

Goffman's total institution

Erving Goffman's Asylums, published in 1961 and based on fieldwork in a psychiatric hospital, supplied the concept that generalises the prison to a family of institutions including monasteries, ships, boarding schools, and army barracks.

A total institution has four features. All aspects of life, sleeping, eating, working, and recreation, occur in one place under one authority. Each phase is carried out in the immediate company of a large batch of others, all treated alike. The day is tightly scheduled by explicit rules and officials. And the activities are organised to fulfil the institution's aims rather than the individual's.

Goffman's sharpest analysis concerns what he called the mortification of the self: the sequence of admission procedures that strips away the equipment a person uses to present an identity. The name replaced by a number. Personal clothing removed and stored. Hair cut. Possessions inventoried. A personal history taken in a form the institution controls. Strip searches. Each step is administratively defensible in isolation, and together they communicate that the self a person brought in is no longer operative.

Goffman's counterweight is equally important. He documented secondary adjustments: the practices by which people in total institutions carve out unauthorised satisfactions and reclaim some autonomy. Trading commissary items. Hidden hobbies. Nicknames. Working the system for a preferred job. These are not merely rule-breaking; they are how people remain persons under conditions designed to prevent it.

The study everyone cites, and why you should not

In August 1971, Philip Zimbardo recruited 24 male college students, assigned them at random to be prisoners or guards, and ran a simulated prison in the basement of the Stanford psychology building. It was terminated after six days. The story that entered every textbook was that ordinary young men, placed in a situation, became brutal guards or broken prisoners within days, and that the situation had overwhelmed disposition.

That interpretation has largely collapsed, and how it collapsed is worth knowing because it is a lesson in reading evidence.

Thibault Le Texier's archival study, published in American Psychologist in 2019, worked through Zimbardo's own recordings and papers held at Stanford. The tapes show that guards were not left to develop their own behaviour: they were briefed and coached toward toughness, and at least one prominent guard later described his behaviour as a deliberate performance. Recordings and later interviews cast doubt on whether the most famous prisoner breakdown was genuine distress or a performance intended to secure release. There was no control condition and no comparison group. The sample was 24 people. And Thomas Carnahan and Sam McFarland demonstrated experimentally that a recruitment advertisement mentioning a study of prison life, as Zimbardo's did, attracts volunteers scoring higher on aggression, authoritarianism, and social dominance than an otherwise identical advertisement without that phrase, so the participants were not a random draw from the population.

There is also a direct empirical rival. Stephen Reicher and Alexander Haslam ran a comparable simulation for the BBC in 2002 with independent ethical oversight and no coaching of guards. Their guards did not spontaneously become tyrannical; the group dynamics that emerged were quite different, and the result was that people do not automatically enact assigned roles.

The point: The Stanford study is not evidence that ordinary people become brutal in prison settings. What we know about prison behaviour comes from ethnography inside real institutions, from administrative records on violence and discipline, and from comparisons between regimes, not from a six-day demonstration with 24 coached participants.

Violence, and how it is measured

The honest state of knowledge about prison violence is that it is undercounted, because reporting requires trusting the institution that would have to record it.

Congress passed the Prison Rape Elimination Act in 2003, which required national data collection and produced the first systematic estimates. The Bureau of Justice Statistics' inmate surveys, which ask people directly and anonymously rather than counting reports, have found on the order of 4 percent of prisoners reporting sexual victimisation in the previous twelve months, with a substantial share involving staff rather than other prisoners. Administrative complaint counts are far lower than survey estimates, which tells you about reporting, not about incidence.

Physical violence follows the same pattern, and it is highly concentrated: most people in most prisons are not assaulted in a given year, and a minority of facilities and units account for a disproportionate share of incidents.

What the courts require

Prisoners retain constitutional rights, and two cases define the modern framework.

Wolff v. McDonnell, decided in 1974, held that a prisoner facing loss of good-time credits in a disciplinary hearing is entitled to advance written notice, a limited opportunity to call witnesses and present evidence, and a written statement of the evidence relied on and the reasons. Not a trial, but not nothing.

Turner v. Safley, decided in 1987, set the standard that governs most prisoner rights claims: a regulation that impinges on a constitutional right is valid if it is reasonably related to legitimate penological interests. That is a deferential test, and courts applying it uphold most challenged regulations.

Congress then narrowed access to court directly. The Prison Litigation Reform Act of 1995 requires prisoners to exhaust internal grievance procedures before filing, bars claims for mental or emotional injury without a prior showing of physical injury, imposes filing fees on indigent prisoners in instalments, and restricts the scope and duration of court orders governing prison conditions. Prisoner civil rights filings per capita fell sharply afterward. Whether that represents the elimination of frivolous suits or the exclusion of meritorious ones is genuinely disputed, and the answer probably includes both.

Common misconceptions

• The Stanford Prison Experiment showed that anyone becomes cruel in a guard role. Guards were coached, participants were self-selected by an advertisement mentioning prison life, there was no control group, and a later independent simulation produced different results.
• Prison officers hold total power. Sykes documented the opposite: officers depend on cooperation and trade informal accommodations for order, so operating rules are negotiated.
• Inmate culture is entirely produced by confinement. The importation model shows that much of it arrives with the population, and the balance between the two mechanisms varies by institution.
• Official complaint counts measure prison violence. Anonymous surveys find rates far above administrative counts, which measure willingness to report to the institution being reported on.
• Prisoners have no rights in disciplinary proceedings. Wolff requires notice, a limited chance to present evidence, and a written statement of reasons, though Turner's deferential standard governs most other claims.

Where this leaves us

• Sykes identified five deprivations, of liberty, goods and services, intimate relationships, autonomy, and security, and treated prison culture as the set of adaptations to them.
• The corruption of authority means officers trade informal tolerance for order, so the real rules of a prison are negotiated rather than imposed.
• Deprivation and importation models both explain part of inmate culture, and their relative weight varies with the institution and its population.
• Goffman's total institution combines batch living under one authority with mortification of the self at admission, met by secondary adjustments that reclaim autonomy.
• The Stanford Prison Experiment does not support the conclusions drawn from it, and the archival and replication evidence against it is substantial.
• Wolff v. McDonnell requires minimal due process in disciplinary hearings, Turner v. Safley applies a deferential reasonableness test, and the 1995 Prison Litigation Reform Act sharply reduced prisoner filings.

Sources

1. Le Texier, T. (2019). Debunking the Stanford Prison Experiment. American Psychologist, 74(7), 823-839. doi.org
2. Wolff v. McDonnell, 418 U.S. 539 (1974). supreme.justia.com
3. Turner v. Safley, 482 U.S. 78 (1987). supreme.justia.com
4. Bureau of Justice Statistics. (n.d.). Prison Rape Elimination Act data collection. Office of Justice Programs. bjs.ojp.gov
5. Sykes, G. M. (1958). The society of captives: A study of a maximum security prison. Princeton, NJ: Princeton University Press.
6. Goffman, E. (1961). Asylums: Essays on the social situation of mental patients and other inmates. New York: Anchor Books.

Twenty years in a windowless cell

In September 2015 the state of California settled Ashker v. Governor of California, agreeing to stop holding people in indefinite isolation on the basis of validated gang affiliation. The litigation record established what the practice had produced at Pelican Bay State Prison: more than 500 men had been held in its Security Housing Unit for over ten years, and 78 of them for more than twenty.

Not twenty years in prison. Twenty years in a cell of roughly 80 square feet, alone, for approximately 22 to 23 hours a day, with an exercise period in a concrete enclosure and no work, no group programming, and no physical contact with another person outside of restraints.

This lesson is about what that does, and about why answering that question is much harder than either side of the public argument admits.

What the practice is called and what it covers

Corrections agencies rarely use the phrase solitary confinement. The operative term is restrictive housing, generally defined as confinement to a cell for 22 hours or more per day with severely limited contact. It covers several distinct legal and administrative situations that get lumped together in public argument and should not be.

Form	Basis	Typical duration	Who decides
Disciplinary segregation	A finding of guilt on a rule violation	Days to months, fixed term	A disciplinary hearing officer
Administrative segregation	A judgement that the person poses a threat to institutional order	Indefinite, subject to periodic review	Classification staff, with limited process
Protective custody	A judgement that the person is at risk from others	Indefinite	Classification staff, sometimes at the person's request
Supermax or SHU placement	Assignment to a whole facility or unit designed for isolation	Years	Central classification, often on gang or assault history

The distinction that matters most is between a fixed disciplinary term and indefinite administrative placement. A person serving thirty days of disciplinary segregation knows the date it ends. A person in administrative segregation on a gang validation may not, and the litigation record shows some of them did not for two decades.

Key idea: Restrictive housing is a family of practices, not one thing. The legal basis, the duration, the review process, and the person deciding all differ, and evidence about one form does not automatically apply to another.

How many people

Nobody knows precisely, because there is no mandatory national reporting and jurisdictions define the category differently. The best available series comes from surveys conducted jointly by the Association of State Correctional Administrators and the Arthur Liman Center at Yale Law School, which asked departments directly. Their mid-2010s estimates put the number in state and federal prisons at roughly 80,000 to 100,000 on a given day. Later rounds reported lower figures for the jurisdictions that responded, reflecting both genuine reductions and changes in which agencies participated. Local jails hold tens of thousands more, and jail practices are the least documented of all.

Treat any single national figure with suspicion. What is not in doubt is the order of magnitude: tens of thousands of people on any given day.

The claimed harms

The clinical case against prolonged isolation is old. In 1890 the Supreme Court, in a case about Colorado's execution procedures, described the observed effects of the earlier solitary systems in terms that would not be out of place in a modern psychiatric text: some prisoners fell into a semi-fatuous condition, others became violently insane, and few recovered enough to be of later use.

The modern clinical description comes from Stuart Grassian's evaluations of prisoners in Massachusetts segregation units in the 1980s. He described a recurring cluster: hypersensitivity to noise and light, perceptual distortions and sometimes hallucinations, panic attacks, difficulty with concentration and memory, intrusive obsessional thoughts, paranoia, and problems with impulse control including self-harm. He argued the cluster was distinctive enough to constitute a syndrome specific to the conditions.

The obvious objection is selection. People placed in segregation are disproportionately people who were already mentally ill, since psychiatric symptoms produce rule violations, and people who are already unwell would look unwell in any setting. Any serious study has to confront that.

Why the evidence is hard, and what substitutes for randomisation

You cannot run the experiment. No ethics board will randomly assign prisoners to two years of isolation, and no department would permit it. So every study of this question uses a design that is weaker than randomisation, and the disputes in the literature are mostly disputes about which weakness matters more.

The strongest study defending the practice is a longitudinal Colorado project led by Maureen O'Keefe, published in 2013. Rather than comparing segregated prisoners with the general population at one moment, it followed people over a year, administering repeated psychological measures to those entering administrative segregation and to comparison groups. Its headline finding was that the segregated group did not deteriorate as the clinical literature predicted, and that psychological measures improved somewhat over time in all groups, including the segregated one.

That result should be taken seriously, and it should be read alongside the criticisms it drew, which are substantial. The comparison groups were not people living in normal conditions; several were themselves in restrictive or highly controlled settings, which compresses the contrast the study was meant to measure. Attrition was considerable, and the people who left the study were plausibly those doing worst. Self-report instruments administered by staff inside a prison to people whose classification status may depend on appearing stable are a difficult measurement problem. And repeated testing itself provides human contact and stimulation, which is the very thing the condition removes.

The strongest studies on the other side use administrative outcomes rather than psychological instruments, which sidesteps the measurement problem entirely.

Fatos Kaba and colleagues linked New York City jail records to health records and examined acts of self-harm. Their finding was stark: solitary confinement was involved in a small share of jail admissions but accounted for roughly half of all acts of self-harm recorded, and a large majority of the potentially fatal ones. Self-harm is recorded because it produces a medical response, so it does not depend on anyone's willingness to report distress.

Lauren Brinkley-Rubinstein and colleagues went further and linked North Carolina prison records to state death records. People who had spent any time in restrictive housing were about 24 percent more likely to die in the first year after release than people who had not, with sharply elevated risks of death by suicide, by homicide, and by opioid overdose, and with the overdose risk concentrated in the first two weeks out. Death certificates are not self-reports, and the follow-up occurs after release, when the institution no longer controls the measurement.

What matters here: The studies that find little harm rely on psychological self-report measured inside the institution. The studies that find substantial harm use recorded self-harm and death records. When designs with very different vulnerabilities point in different directions, the honest reading gives more weight to outcomes that cannot be shaped by the setting in which they are measured.

What the law requires

The constitutional position is narrower than most people expect.

In Hutto v. Finney, decided in 1978, the Supreme Court upheld a lower court's order limiting punitive isolation in the Arkansas system to thirty days, as a remedy for conditions found unconstitutional as a whole. That is a remedial ceiling in one case, not a general rule.

In Wilkinson v. Austin, decided in 2005, the Court held that assignment to Ohio's supermax facility, given its severity and indefinite duration, implicates a protected liberty interest, so some process is due. The process the Court approved was informal: notice of the grounds, an opportunity to respond in writing, and periodic review. Not a hearing with counsel and witnesses.

Litigation has moved further where mental illness is involved. Federal courts, beginning with the Pelican Bay litigation in the 1990s, have held that placing people with serious mental illness in these conditions can violate the Eighth Amendment, and many states now formally exclude such prisoners, though implementation varies.

Internationally, the United Nations Standard Minimum Rules for the Treatment of Prisoners, adopted in 2015 and known as the Mandela Rules, define prolonged solitary confinement as more than fifteen consecutive days and prohibit both prolonged and indefinite solitary confinement. American practice at Pelican Bay exceeded that limit by a factor of roughly five hundred.

The administrators' case, stated properly

The argument for retaining segregation is not usually made well by its critics, so here it is at full strength.

A prison holds people who have assaulted staff and other prisoners, some of whom will do so again. A warden who removes the ability to separate such a person has to manage them somewhere, and the alternatives, small structured units with high staffing ratios, intensive mental health treatment, and step-down programming, require money and personnel that legislatures have not appropriated. Eliminating segregation without funding the alternative does not make the risk disappear; it transfers it onto other prisoners and onto officers, who are themselves assaulted at meaningful rates. On this view segregation is a bad tool that exists because the good tools were never bought.

The best answer to that argument is empirical rather than rhetorical, and it comes from states that made the change and measured what happened. Mississippi reduced the population of its notorious Unit 32 from around a thousand people to roughly a hundred and fifty, with violence falling rather than rising. Colorado under director Rick Raemisch cut administrative segregation drastically and reported no resulting increase in institutional violence. North Dakota rebuilt its practice around short, purposeful placements with structured return to population. Those are real reductions with measured outcomes, and they suggest that a substantial share of segregation placements were not managing risk at all.

Common misconceptions

• Solitary confinement is one practice. Disciplinary, administrative, protective, and supermax placements differ in basis, duration, and process, and evidence about one does not transfer to another.
• The research shows isolation is harmless. One longitudinal study using in-prison self-report found little deterioration; studies using recorded self-harm and death records find substantial harm.
• The Constitution limits solitary to thirty days. Hutto approved a thirty-day cap as a remedy in one case; there is no general constitutional limit, and Wilkinson requires only informal process.
• Reducing segregation increases prison violence. Mississippi, Colorado, and North Dakota made large reductions without increases in violence, which is the strongest evidence against the operational case.
• Everyone in restrictive housing was placed there for violence. Administrative placements have included gang validation based on association and possession of literature, which is what the Ashker settlement addressed.

What to remember

• The Ashker litigation established that over 500 men at Pelican Bay had been held in isolation for more than ten years and 78 for more than twenty.
• Restrictive housing means 22 or more hours a day in cell, and covers disciplinary, administrative, protective, and supermax placements with different bases and durations.
• The best national estimates, from the ASCA and Liman surveys, put the mid-2010s prison figure in the region of 80,000 to 100,000 on a given day, with jails poorly documented.
• No randomised evidence exists or could exist, so every study substitutes a weaker design, and the disputes are about which weakness matters most.
• The Colorado longitudinal study found little psychological deterioration using in-prison self-report; studies of recorded self-harm in New York City jails and of post-release mortality in North Carolina found substantial harm using measures the institution does not control.
• The Mandela Rules define prolonged solitary as beyond fifteen consecutive days and prohibit it, while American constitutional law requires only informal process for indefinite supermax placement.

Sources

1. Kaba, F., Lewis, A., Glowa-Kollisch, S., Hadler, J., Lee, D., Alper, H., Selling, D., MacDonald, R., Solimo, A., Parsons, A., and Venters, H. (2014). Solitary confinement and risk of self-harm among jail inmates. American Journal of Public Health, 104(3), 442-447. doi.org
2. Brinkley-Rubinstein, L., Sivaraman, J., Rosen, D. L., Cloud, D. H., Junker, G., Proescholdbell, S., Shanahan, M. E., and Ranapurwala, S. I. (2019). Association of restrictive housing during incarceration with mortality after release. JAMA Network Open, 2(10), e1912516. doi.org
3. Wilkinson v. Austin, 545 U.S. 209 (2005). supreme.justia.com
4. Hutto v. Finney, 437 U.S. 678 (1978). supreme.justia.com
5. National Institute of Justice. (n.d.). Corrections research, including restrictive housing. Office of Justice Programs. nij.ojp.gov
6. United Nations Office on Drugs and Crime. (2015). The United Nations standard minimum rules for the treatment of prisoners: The Nelson Mandela Rules. unodc.org

A death every six or seven days

When the Supreme Court decided Brown v. Plata in May 2011, it was reviewing an order issued by a three-judge federal court that had spent years on the California prison system. The record established that the system was operating at roughly double its design capacity, that prisoners with serious mental illness were being held in telephone-booth-sized cages awaiting treatment beds, that suicides were occurring at rates well above the national prison average, and that a prisoner in California was dying needlessly, of a preventable cause, on the order of once every six or seven days.

The three-judge court had ordered the state to reduce its prison population to 137.5 percent of design capacity, on the reasoning that no amount of money spent on medical staffing could deliver adequate care in a system that crowded. The Supreme Court affirmed, by five votes to four.

That is where American correctional health care law sits, and it took thirty-five years of litigation to get there.

The only constitutional right to health care in America

Start with the doctrinal foundation, because it produces a genuinely strange fact about the United States.

In Estelle v. Gamble, decided in 1976, the Court held that deliberate indifference to serious medical needs of prisoners violates the Eighth Amendment's prohibition on cruel and unusual punishment. The reasoning was straightforward: the state has deprived a person of the ability to obtain care for themselves, so failing to provide it inflicts unnecessary suffering that serves no penological purpose.

The consequence is that incarcerated people are the only group in the United States with a constitutional right to health care. Nobody outside a correctional facility has one.

The standard is demanding, though, and students consistently overestimate it. It has two parts. The medical need must be objectively serious. And the official must be subjectively indifferent, which Farmer v. Brennan defined in 1994 as knowing of and disregarding an excessive risk to health or safety. Negligence does not meet it. Malpractice does not meet it. A doctor who misdiagnoses a condition through carelessness has not violated the Constitution; a nurse who reads a chart, sees a critical result, and files it without acting may have.

Key idea: Estelle creates a right to care but sets the constitutional floor at deliberate indifference, not at competent medicine. The gap between the floor and adequate care is where most of the actual problem lives, and it is not reachable by constitutional litigation.

Who is inside, medically speaking

Correctional populations are far sicker than the general population, for reasons that precede custody. People entering jail and prison are drawn disproportionately from groups with low incomes, unstable housing, and little prior access to care, and they carry the accumulated consequences.

Rates of chronic conditions such as hypertension, asthma, and diabetes run above general population rates at equivalent ages. Rates of infectious disease, particularly hepatitis C and HIV, run well above. Substance use disorder is present in a large fraction of the population. And mental illness is present at rates that make correctional facilities, in practice, the country's largest psychiatric providers.

The Bureau of Justice Statistics has surveyed this directly. In its inmate surveys, roughly 37 percent of state and federal prisoners and 44 percent of people in jail reported having been told by a mental health professional that they had a mental disorder. Using a standardised screening instrument for the previous 30 days, around 14 percent of prisoners and 26 percent of jail inmates met the threshold for serious psychological distress. Compare the jail figure with general population estimates, which run in the low single digits, and the scale of the concentration is obvious.

By raw headcount, the largest facilities holding people with serious mental illness in the United States are county jails, not hospitals.

The transinstitutionalisation argument, and its problem

The standard explanation for that concentration is compact and satisfying, and it is only partly right.

The story goes: America emptied its state psychiatric hospitals, whose resident census fell from roughly 560,000 in the mid-1950s to a small fraction of that; the community mental health system that was supposed to replace them was never adequately funded; and the people who would once have been hospitalised ended up in jails and prisons instead. On this account, one institution simply replaced another, a process called transinstitutionalisation.

The problem is timing. The great majority of the hospital population decline occurred between the mid-1950s and about 1980. The great majority of prison population growth occurred after 1980. If the same people were being moved from one institution to the other, the curves should overlap more than they do.

The careful position, which is what the research literature actually supports, is that the relationship is real but indirect and partial. Deinstitutionalisation without adequate community services created a population of people with serious mental illness living in unstable circumstances. Policy changes decades later, including aggressive enforcement of low-level offences, drug criminalisation, and the withdrawal of housing and social supports, then swept a large share of that population into jails. The mechanism runs through the absence of community treatment rather than through a direct transfer of patients.

That distinction matters for policy. If the problem were a simple transfer, rebuilding hospitals would fix it. If the problem is the absence of community treatment interacting with enforcement practice, then building hospital beds without changing either would not.

The most dangerous fortnight

Here is the single most striking finding in correctional health research, and it concerns the period immediately after release.

Ingrid Binswanger and colleagues linked Washington State prison records to death records and compared released prisoners with other state residents of the same age, sex, and race. In the first two weeks after release, the mortality rate among former prisoners was nearly thirteen times higher than the comparison rate. The leading cause was drug overdose.

The mechanism is physiological and brutally simple. Opioid tolerance declines during a period of enforced abstinence in custody. A person who resumes using at the dose that was normal before incarceration takes a quantity their body can no longer handle. Release is therefore not a return to baseline risk; it is a spike.

Knowing the mechanism tells you the intervention. Medications for opioid use disorder, meaning methadone, buprenorphine, and naltrexone, prevent that spike, and the evidence for them in community settings is among the strongest in addiction medicine. Most American correctional systems historically did not provide them, on the reasoning that abstinence was the goal and that the medications were diverted.

Rhode Island tested the alternative at state scale. Beginning in 2016 it screened everyone entering its unified corrections system and made all three medications available, continuing them through release into community treatment. Traci Green and colleagues compared overdose deaths among recently incarcerated people before and after implementation and found a large reduction, on the order of 60 percent, at a time when overdose deaths in the state's general population were not falling comparably. It is a single state with a before-and-after design rather than a randomised one, and it is the clearest natural experiment available.

What matters here: The two weeks after release carry roughly thirteen times the general mortality risk, driven by overdose from lost tolerance. Providing medications for opioid use disorder in custody and continuing them at release addresses the mechanism directly, and the Rhode Island result suggests the effect is large.

Structural problems that litigation cannot reach

Four features of the system produce poor care independently of anyone's indifference.

The Medicaid inmate exclusion. Federal Medicaid dollars generally cannot be used for care provided to people who are inmates of a public institution. States therefore bear the full cost of correctional health care from their own budgets, and enrolment is typically terminated or suspended at booking, so people leave custody uninsured at the exact moment their risk peaks. Recent federal waivers permitting limited pre-release services are an attempt to close that gap.

Delivery structure. Correctional health care is provided variously by state employees, by contracts with university medical schools, and by private vendors paid on a capitated basis. A capitated contract pays a fixed amount per prisoner regardless of care delivered, which places the financial incentive on the wrong side of every treatment decision.

Co-payments. Many systems charge a fee, often small in absolute terms, for a medical visit. In a setting where wages are measured in cents per hour, a few dollars is a real price, and it deters the early presentation that keeps conditions cheap and survivable.

Ageing. The share of state prisoners aged 55 and over has multiplied since the early 1990s, a direct consequence of long sentences imposed decades earlier. Older prisoners require the most care, cost several times more to hold, and have the lowest reoffending rates in the system.

Common misconceptions

• Estelle guarantees prisoners good medical care. It prohibits deliberate indifference to serious needs, which is a much lower floor; malpractice and negligence do not violate it.
• Brown v. Plata was about medical staffing budgets. The three-judge court concluded that no staffing level could deliver adequate care at roughly double design capacity, so the remedy was a population cap.
• Emptying psychiatric hospitals directly filled prisons. The hospital decline was largely complete by 1980 and prison growth came after, so the link runs through absent community treatment interacting with later enforcement policy.
• Leaving prison reduces health risk immediately. The two weeks after release carry roughly thirteen times the comparable mortality risk, driven by overdose from reduced opioid tolerance.
• Medication for opioid use disorder is unavailable because it does not work in custody. It works; the barriers have been policy preferences for abstinence and concerns about diversion, and Rhode Island's implementation was followed by a large fall in post-release overdose deaths.

Pulling it together

• Estelle v. Gamble established that deliberate indifference to serious medical needs violates the Eighth Amendment, making prisoners the only Americans with a constitutional right to health care.
• Farmer v. Brennan defined the subjective element as knowing of and disregarding an excessive risk, so negligence and malpractice fall below the constitutional floor.
• Brown v. Plata upheld a population cap at 137.5 percent of design capacity on a record including a needless death roughly every six or seven days.
• Roughly 37 percent of prisoners and 44 percent of jail inmates report a prior mental disorder diagnosis, and jails are by headcount the largest facilities holding people with serious mental illness.
• Transinstitutionalisation is real but indirect: hospital closure preceded prison growth by decades, and the link runs through absent community treatment plus later enforcement policy.
• Mortality in the first two weeks after release runs about thirteen times the comparison rate, and Rhode Island's provision of all three opioid use disorder medications was followed by roughly a 60 percent fall in overdose deaths among recently incarcerated people.

Sources

1. Estelle v. Gamble, 429 U.S. 97 (1976). supreme.justia.com
2. Brown v. Plata, 563 U.S. 493 (2011). supreme.justia.com
3. Binswanger, I. A., Stern, M. F., Deyo, R. A., Heagerty, P. J., Cheadle, A., Elmore, J. G., and Koepsell, T. D. (2007). Release from prison: A high risk of death for former inmates. New England Journal of Medicine, 356(2), 157-165. doi.org
4. Green, T. C., Clarke, J., Brinkley-Rubinstein, L., Marshall, B. D. L., Alexander-Scott, N., Boss, R., and Rich, J. D. (2018). Postincarceration fatal overdoses after implementing medications for addiction treatment in a statewide correctional system. JAMA Psychiatry, 75(4), 405-407. doi.org
5. Bureau of Justice Statistics. (n.d.). Health of prisoners and jail inmates. Office of Justice Programs. bjs.ojp.gov
6. Substance Abuse and Mental Health Services Administration. (n.d.). Medications for substance use disorders. samhsa.gov

Forty-three percent lower odds

In 2013 a RAND team led by Lois Davis published a meta-analysis pooling evaluations of correctional education across three decades of American studies. People who participated in an education programme while incarcerated had roughly 43 percent lower odds of returning to prison than comparable non-participants, and roughly 13 percent higher odds of obtaining employment after release.

Those numbers are widely quoted and widely misunderstood, in two separate ways that this lesson will fix. The first misunderstanding is about what an odds reduction means numerically. The second is about what a meta-analysis of mostly non-experimental studies can establish about cause. Both matter, and getting them right is most of what it takes to read this literature.

The framework that organises everything: risk, need, responsivity

Before individual programmes, the model that predicts which ones will work. Don Andrews and James Bonta developed it from decades of meta-analytic work, and it has three principles.

Risk. Match the intensity of service to the person's assessed risk of reoffending. Intensive programming should go to high-risk people. This is not a claim about who deserves help; it is a claim about where the measurable reduction is available, since a person unlikely to reoffend has little room to improve.

Need. Target criminogenic needs, meaning changeable characteristics that actually predict offending, rather than characteristics that are merely unpleasant. Andrews and Bonta's central eight are antisocial cognition, antisocial associates, an antisocial personality pattern, a history of antisocial behaviour, family and relationship problems, difficulties with school and work, unstructured leisure, and substance abuse. Self-esteem, notably, is not among them; programmes that raise self-esteem without addressing the eight do not reduce offending.

Responsivity. Deliver services in ways people can actually absorb, which the evidence says means structured cognitive and social learning methods, and adapt to individual factors such as literacy, motivation, and mental health.

The risk principle has an implication that startles people the first time they meet it, and it is well supported. Placing low-risk people into intensive programming can increase their recidivism. Christopher Lowenkamp and Edward Latessa documented this pattern across Ohio programmes. The mechanisms are plausible: intensive programming pulls a person away from the job, family, and routine that were keeping them out of trouble, and it places them in sustained contact with higher-risk peers who supply exactly the antisocial associates the model identifies as criminogenic.

Key idea: More treatment is not better treatment. The risk principle says intensive services help high-risk people and can harm low-risk people, which means that a programme's average effect depends on who was put into it.

Reading an effect size without fooling yourself

Now the arithmetic, because this is where most public discussion of correctional programming goes wrong.

Suppose a programme reduces recidivism by 25 percent. That is a relative reduction, and it must be applied to a base rate. If the comparison group reoffends at 50 percent, a 25 percent relative reduction produces 37.5 percent, not 25 percent. The programme moved one person in eight, not one in four.

The same trap sits inside odds ratios. The RAND figure is 43 percent lower odds, and odds are not probabilities. When the underlying event is common, an odds ratio overstates the change in probability. A 43 percent reduction in odds from a base recidivism rate near 50 percent corresponds to a probability of roughly 36 percent, an absolute change of about fourteen percentage points. That is a substantial and worthwhile effect. It is not the near-halving that the phrase 43 percent suggests to a casual reader.

Get in the habit of asking three questions of any reported effect. What was the base rate? Is this relative or absolute? And is it odds or probability? A great deal of correctional policy advocacy survives only because those questions are not asked.

What the evidence supports

Cognitive behavioural programmes have the strongest evidence base in the field. Nana Landenberger and Mark Lipsey's meta-analysis of controlled evaluations found a mean recidivism reduction of roughly 25 percent, with the largest effects for higher-risk participants, for programmes implemented with fidelity, and for curricula including anger control and interpersonal problem solving components. The mechanism targets a criminogenic need directly: the thinking patterns that make offending seem reasonable in the moment.

Correctional education shows the RAND result above. Its weakness is design. Most of the pooled studies compare participants with non-participants, and people who enrol in and complete education programmes differ from those who do not in motivation, institutional conduct, and sentence structure. RAND was explicit about this limitation. The finding is consistent, large, and confounded, all at once. Note also the policy shift: after a pilot programme beginning in 2015, full Pell Grant eligibility for incarcerated students was restored effective in 2023, which will over time generate better evidence simply by increasing the number of programmes to evaluate.

Substance use treatment shows a pattern worth memorising because it recurs. In-prison therapeutic communities, which are intensive residential programmes separating participants from the general population, produce modest reductions in reoffending on their own. Combined with continued treatment after release, the reductions are substantially larger. The Delaware evaluations of the Key and Crest programmes are the standard demonstration. The lesson generalises: an in-prison programme that ends at the gate is competing with the entire environment a person returns to.

Prison work and industries show favourable associations with post-release employment and lower recidivism, with the same selection problem as education, and with an additional complication: assignment to desirable work is itself a reward for good conduct, so the comparison is between people the institution already judged differently.

What does not work, and what makes things worse

A field that only reported successes would be useless. Three well-evidenced failures follow.

Scared Straight programmes bring young people into prisons for confrontational sessions with prisoners describing the realities of confinement. The intuition is obvious and the finding is the reverse. Anthony Petrosino and colleagues, in a Campbell Collaboration systematic review of randomised trials, found that participants offended more than controls, with pooled odds of offending substantially above one. This is one of the clearest demonstrations in criminology that a plausible intervention can cause harm, and that intuition is not evidence.

Correctional boot camps, with military drill, discipline, and physical training, have been evaluated repeatedly. Doris MacKenzie's syntheses find no reliable effect on recidivism. Where individual camps showed effects, they were attributable to the treatment components some of them included, not to the drill.

Intensive supervision without treatment is the third. A RAND experiment led by Joan Petersilia and Susan Turner randomly assigned probationers and parolees across fourteen sites to intensive supervision or routine supervision. Intensive supervision did not reduce new arrests. It substantially increased detected technical violations and therefore revocations, because closer watching finds more infractions. Surveillance without services increased incarceration without reducing crime.

The point: Programmes are not inert. Scared Straight increases offending, and intensive surveillance increases incarceration without reducing crime. An evaluation that reports no effect is good news compared with the alternatives that exist.

Implementation is not a footnote

A programme in a manual and a programme on a tier are different objects, and the gap between them explains a large share of the variation in results.

Instruments such as the Correctional Program Assessment Inventory score real programmes on the things the research says matter: whether staff are trained and supervised, whether participants are assessed and matched by risk, whether the curriculum is delivered as designed for the full dosage, and whether the programme is evaluated at all. When these instruments are applied in the field, a large share of programmes score poorly. An evaluation of a well-designed programme delivered badly measures the delivery, not the design.

This is why practitioners should read the CrimeSolutions ratings maintained by the National Institute of Justice, and why researchers report implementation fidelity alongside outcomes. It is also why the sentence a programme reduces recidivism by X percent is incomplete without the phrase when delivered as designed to the population it was designed for.

Money, and the case that persuades legislatures

The Washington State Institute for Public Policy pioneered a form of analysis that has done more to change correctional practice than most published research: systematic benefit-cost estimates in which each programme's measured effect on recidivism is converted into avoided criminal justice costs and avoided victimisation costs, and set against the programme's price.

The rankings that come out of this are informative in a way that effect sizes alone are not. A programme with a modest effect that costs very little can outrank one with a larger effect that costs a great deal, because what a legislature actually chooses between is programmes at a given budget rather than programmes in the abstract.

Common misconceptions

• Everyone in prison should receive intensive programming. The risk principle holds that intensive services help high-risk people and can increase offending among low-risk people.
• A 43 percent reduction means recidivism falls by nearly half. That figure is an odds reduction, which from a base rate near 50 percent corresponds to roughly a fourteen percentage point absolute change.
• Confronting young people with prison reality deters them. Randomised trials pooled by the Campbell Collaboration find Scared Straight participants offend more than controls.
• Closer supervision reduces crime. The RAND intensive supervision experiment found no reduction in new arrests and a large increase in detected technical violations and revocations.
• A programme with good evidence will work anywhere. Implementation fidelity varies enormously, and an evaluation of a well-designed programme delivered badly measures the delivery.

What you now know

• The RAND meta-analysis found roughly 43 percent lower odds of return to prison and 13 percent higher odds of employment among participants in correctional education, on mostly non-experimental designs.
• Risk, need, and responsivity together predict which programmes work: intensity matched to risk, targeting of criminogenic needs, and structured cognitive social learning delivery.
• The risk principle predicts, and evidence confirms, that intensive programming for low-risk people can increase their offending.
• Cognitive behavioural programmes reduce recidivism by roughly 25 percent on average, with larger effects for higher-risk participants and faithful implementation.
• In-prison therapeutic communities produce modest reductions alone and substantially larger ones when treatment continues after release.
• Scared Straight increases offending, boot camps show no reliable effect, and intensive supervision without treatment raises revocations without reducing new arrests.

Sources

1. Davis, L. M., Bozick, R., Steele, J. L., Saunders, J., and Miles, J. N. V. (2013). Evaluating the effectiveness of correctional education: A meta-analysis of programs that provide education to incarcerated adults. Santa Monica, CA: RAND Corporation. rand.org
2. National Institute of Justice. (n.d.). CrimeSolutions: Programme and practice ratings. Office of Justice Programs. crimesolutions.ojp.gov
3. Petrosino, A., Turpin-Petrosino, C., Hollis-Peel, M. E., and Lavenberg, J. G. (2013). Scared Straight and other juvenile awareness programs for preventing juvenile delinquency. Campbell Systematic Reviews. campbellcollaboration.org
4. Landenberger, N. A., and Lipsey, M. W. (2005). The positive effects of cognitive-behavioral programs for offenders: A meta-analysis of factors associated with effective treatment. Journal of Experimental Criminology, 1(4), 451-476. doi.org
5. Washington State Institute for Public Policy. (n.d.). Benefit-cost results for adult criminal justice programs. wsipp.wa.gov
6. Andrews, D. A., and Bonta, J. (2010). The psychology of criminal conduct (5th ed.). New Providence, NJ: LexisNexis Anderson.

Forty-five percent of admissions, and nobody was arrested for a new crime

In 2019 the Council of State Governments Justice Center published a state-by-state accounting of why people enter prison. Nationally, 45 percent of state prison admissions were the result of violations of probation or parole rather than new convictions. Roughly one in four of all admissions was for a technical violation alone: missing appointments, failing a drug test, moving without permission, not paying supervision fees. No new crime was charged.

That finding reorganises how you should think about American incarceration. A large share of the people entering prison in a given year are not being sentenced for an offence. They are failing at supervision, which is a different problem with different solutions.

What probation is, and what it is not

Get the distinction right first, because it is the single most common error in this area.

Probation is a sentence. A court convicts a person and, instead of imposing confinement, orders them to live in the community under conditions and supervision, often with a suspended prison term hanging over them if they fail.

Parole is a release. A person who is already serving a prison sentence is released before its expiration and serves the remainder in the community under supervision.

The person supervising them may work for the same agency and their day-to-day experience may look identical, but the legal posture is different: the probationer has not been imprisoned and the parolee has.

Probation is by a wide margin the most common criminal sentence in the United States. Something close to three million adults are on probation at any moment, alongside roughly 800,000 on parole. Together, community supervision holds more than twice as many people as all prisons and jails combined.

The practice has a specific origin. John Augustus, a Boston bootmaker, began in 1841 posting bail for defendants he thought could be helped, supervising them himself, and returning to court with reports. He did this for thousands of people over eighteen years, at his own expense. Massachusetts enacted the first state probation statute in 1878.

Key idea: Probation is a sentence served in the community and parole is early release from a prison sentence. Both are supervision, and together they hold most of the people under correctional control in the United States.

The conditions, and how they accumulate

A supervision order is a list of rules. Standard conditions typically include reporting as directed, permitting home visits, not leaving the jurisdiction without permission, not possessing firearms, maintaining employment, not associating with people who have criminal records, and abstaining from drugs and alcohol. Special conditions are added by the court: treatment programmes, curfews, electronic monitoring, no-contact orders, restitution, community service.

They accumulate. It is entirely ordinary for a person to be under fifteen or twenty simultaneous obligations, several of which are in tension. Maintain employment and attend a treatment programme that meets during working hours. Do not associate with anyone who has a record, in a neighbourhood where many people do. Do not leave the county, and get to a job in the next county.

Then there is money. Many jurisdictions run supervision on an offender-funded basis: the supervised person pays a monthly supervision fee, pays for their own drug tests, pays for the electronic monitoring device and its daily charge, and pays court costs, fines, and restitution. Where private companies provide misdemeanour probation, the company's revenue comes directly from those fees. Non-payment is itself a violation.

The design problem should be obvious. A system that supervises a population selected for poverty, and funds itself by charging them, has built a failure mechanism into its financing.

Technical violations and revocation

Violations come in two kinds and they are not comparable.

A new offence violation means the person has been arrested or convicted of a new crime. A technical violation means they have broken a supervision rule that is not itself a crime.

Revocation returns the person to custody, and the procedural protections are limited. In Gagnon v. Scarpelli, decided in 1973, the Supreme Court held that probation revocation requires a preliminary hearing on probable cause and a final hearing with notice, disclosure of the evidence, an opportunity to be heard and to present witnesses, a neutral decision-maker, and a written statement of reasons. It declined to require appointed counsel in every case, leaving that to be decided case by case.

Note what the standard of proof is not. Revocation typically proceeds on a preponderance of the evidence before a judge or hearing officer, not beyond a reasonable doubt before a jury. A person can be imprisoned on a revocation for conduct that could not have sustained a criminal conviction.

The officer's impossible job

A probation officer is asked to do two things that pull in opposite directions.

The law enforcement role is surveillance and enforcement: verify compliance, test for drugs, conduct home visits, and report violations. The social work role is assistance: connect people to treatment, housing, and employment, and help them succeed.

These conflict at the level of information. A person who discloses a relapse to an officer they trust has just handed that officer grounds for a violation report. Every honest conversation is potential evidence, which is a poor foundation for a helping relationship.

Caseloads make it worse. Typical American probation caseloads run well over a hundred people per officer and in some jurisdictions above two hundred. At that ratio, meaningful individual attention is arithmetically impossible, and what remains is processing: a brief monthly check-in and a urine sample.

HOPE, and a lesson about replication

The most instructive episode in the supervision literature concerns a programme that appeared to solve the whole problem.

In 2004 Judge Steven Alm in Hawaii began an experiment called HOPE, Hawaii's Opportunity Probation with Enforcement. The theory came straight from the deterrence literature: certainty matters more than severity. Every violation, however minor, would receive an immediate and certain sanction, but a modest one, typically a few days in jail rather than revocation of the whole probation term. Participants were warned in advance, in person, exactly what would happen.

The initial randomised evaluation by Angela Hawken and Mark Kleiman was remarkable. Compared with probationers on standard supervision, HOPE participants had far fewer positive drug tests, missed far fewer appointments, were arrested less often for new crimes, and had their probation revoked less often. The findings were widely publicised, and legislatures in several states enacted HOPE-style programmes.

Then came the replication. A federally funded Demonstration Field Experiment implemented the model in four sites in different states, with random assignment. Across those sites, HOPE-style supervision produced no significant reductions in new arrests, and revocations were not consistently lower. In some sites participants were more likely to be jailed, since the model requires frequent short jail stays by design.

Why the difference? Several candidate explanations, all instructive. Hawaii's original programme was run by an unusually committed judge with strong buy-in from prosecutors, defenders, and police, which is a condition a multi-site rollout cannot reproduce. The comparison condition differed: Hawaii's standard probation was particularly unresponsive, so the contrast was larger. And the swiftness the model requires depends on operational capacity, warrants served immediately, hearings within days, jail beds available, that the replication sites did not consistently have.

What matters here: A striking result from a single site with an exceptional champion is a hypothesis, not a finding. HOPE is the field's clearest case, and the correct response is neither to dismiss the original nor to keep citing it without the replication.

What the evidence does support

Several things about supervision are better established.

Surveillance without services does not reduce crime. The RAND intensive supervision experiment covered in the previous lesson found no reduction in new arrests and a large increase in detected technical violations. Adding treatment to intensive supervision does change outcomes; adding only monitoring does not.

Most failure happens early. Violations and new arrests are heavily concentrated in the first months of a supervision term. Terms of five or ten years therefore consume officer capacity supervising people whose risk period has largely passed, which is the empirical argument behind earned early termination policies.

Reducing conditions and caseloads together shows promise. Several states restructured supervision under justice reinvestment programmes, capping the incarceration time available for technical violations, introducing graduated administrative sanctions that do not require a court hearing, and giving counties financial incentives to reduce revocations. Arizona's legislation of 2008 is a documented example, with substantial reported reductions in revocations to prison alongside falling new felony convictions among probationers.

Electronic monitoring may substitute for custody more effectively than it supplements supervision. Rafael Di Tella and Ernesto Schargrodsky exploited the effectively random assignment of Argentine detainees to judges who differed in their use of electronic monitoring, and found substantially lower recidivism among those monitored at home than among those held in prison. That is a comparison with incarceration, not with ordinary probation, and the distinction is exactly the one policymakers blur.

Common misconceptions

• Probation and parole are the same thing. Probation is a sentence imposed instead of confinement; parole is release from a prison sentence already being served.
• People go to prison for committing crimes. Nationally, 45 percent of state prison admissions follow supervision violations, and about a quarter of all admissions are for technical violations alone.
• Revocation requires proof beyond a reasonable doubt. It typically proceeds on a preponderance before a hearing officer, with the limited process Gagnon requires and no automatic right to counsel.
• HOPE probation is proven to work. The Hawaii result was striking and the four-site randomised replication found no significant reductions in new arrests.
• Electronic monitoring is a proven alternative to probation. The strongest evidence compares monitoring with imprisonment, not with ordinary supervision, and those are different questions.

Summing up

• Forty-five percent of state prison admissions follow supervision violations, and roughly one in four admissions is for a technical violation with no new charge.
• Probation is a sentence and parole is a release; together they supervise more than twice as many people as prisons and jails hold.
• Supervision conditions accumulate into fifteen or twenty simultaneous obligations, and offender-funded systems charge the supervised population fees whose non-payment is itself a violation.
• Gagnon v. Scarpelli requires a preliminary and a final hearing with limited process, and revocation runs on a preponderance standard rather than proof beyond a reasonable doubt.
• The HOPE model produced a striking single-site randomised result and no significant reductions in new arrests across a four-site randomised replication.
• Failure is concentrated early in supervision terms, surveillance without services does not reduce new crime, and the strongest electronic monitoring evidence compares it with imprisonment rather than with probation.

Sources

1. Council of State Governments Justice Center. (2019). Confined and costly: How supervision violations are filling prisons and burdening budgets. csgjusticecenter.org
2. Bureau of Justice Statistics. (n.d.). Probation and parole in the United States. Office of Justice Programs. bjs.ojp.gov
3. Gagnon v. Scarpelli, 411 U.S. 778 (1973). supreme.justia.com
4. Di Tella, R., and Schargrodsky, E. (2013). Criminal recidivism after prison and electronic monitoring. Journal of Political Economy, 121(1), 28-73. doi.org
5. National Institute of Justice. (n.d.). Community corrections research. Office of Justice Programs. nij.ojp.gov

Revoked by memorandum

John Morrissey was released on parole from an Iowa prison in June 1967. Seven months later his parole officer filed a report stating that he had bought a car under an assumed name, driven it without permission, obtained credit by giving false information, and given a false account of a car accident. On the strength of that report, and without any hearing, the parole board revoked his parole. He was returned to prison to serve the balance of his sentence.

In 1972 the Supreme Court held in Morrissey v. Brewer that this violated the Due Process Clause. Parole revocation, the Court said, involves a loss of conditional liberty that is real even if it is not the full liberty of an unconvicted person, and it requires process: a prompt preliminary hearing near the place of arrest to determine probable cause, and then a final hearing with written notice of the claimed violations, disclosure of the evidence, an opportunity to be heard and to present witnesses, the chance to confront adverse witnesses in most circumstances, a neutral hearing body, and a written statement of the evidence relied on and the reasons.

Note what the Court did not do. It did not require a jury, proof beyond a reasonable doubt, or appointed counsel as a matter of course. Parole procedure is a floor, not a trial.

Three ways to leave prison

People leave prison in three legally distinct ways, and the mix has shifted enormously.

Discretionary parole is release by a decision of a parole board before the sentence expires, on the board's judgement about readiness and risk. Mandatory release is release on a date computed from the sentence and accumulated good-time credits, with no discretionary judgement involved. Expiration is serving the full term, after which supervision may not follow at all.

In the 1970s the great majority of American prison releases were discretionary parole board decisions. The determinate sentencing turn described in this course's history lesson abolished or narrowed parole boards in many jurisdictions, including the federal system, and by the 2000s most releases nationally were mandatory. Discretionary parole remains dominant in a substantial minority of states, and the difference matters: in a discretionary state, a sentencing judge sets an outer bound and a board sets the real date; in a determinate state, the judge and the good-time statute set it together, and the board plays no role.

The constitutional question of whether anyone is entitled to release was settled early. In Greenholtz v. Inmates of the Nebraska Penal and Correctional Complex, decided in 1979, the Court held that there is no constitutional right to be released before the expiration of a valid sentence. A state may create a liberty interest by writing its statute in mandatory terms, in which case some process is due, but a discretionary parole statute creates no such entitlement.

Key idea: Morrissey guarantees process when parole is taken away, and Greenholtz denies any entitlement to parole in the first place. The asymmetry is deliberate: losing conditional liberty is a deprivation, while not receiving it is not.

How boards actually decide, and a cautionary study

Parole boards are typically small panels of gubernatorial appointees. In many states there is no statutory requirement of professional expertise in corrections, psychology, or criminology. Hearings are often brief. Reasons for denial are frequently given as a checkbox referring to the seriousness of the offence, which is a fact that will never change and therefore supplies a ground for denial that can be repeated indefinitely.

The most cited empirical study of this decision-making illustrates something else, and it is worth working through carefully.

In 2011 Shai Danziger, Jonathan Levav, and Liora Avnaim-Pesso published an analysis in PNAS of over a thousand parole decisions by Israeli judges. Their finding was startling: the proportion of favourable rulings started high after each food break, around 65 percent, declined steadily across the session to nearly zero, and reset abruptly after the next break. The result travelled worldwide as evidence that judicial decisions depend on when the decision-maker last ate.

The critique arrived quickly and is substantial. Keren Weinshall-Margel and John Shapard, working with knowledge of Israeli court practice, established that cases were not ordered randomly. Prisoners with legal representation were generally scheduled earlier in a session, and cases from a single prison were heard together, so the composition of the queue changed systematically across the session. Others noted that the reported effect size, from roughly two thirds favourable to essentially none, is implausibly large for any glucose-based mechanism and would imply an effect far greater than anything demonstrated in controlled work on decision fatigue.

The honest summary is that the study documented a real pattern in the data and that the causal interpretation attached to it does not survive scrutiny. It remains one of the most cited findings in behavioural science, and the critique is almost never cited alongside it.

From clinical judgement to actuarial prediction

The alternative to intuition is statistics, and the argument for it is old. Paul Meehl's 1954 review compared clinical prediction, in which an expert weighs information and forms a judgement, with actuarial prediction, in which the same information is combined by a fixed formula. Across a large number of comparisons, the formula did as well or better. That result has been replicated many times in many fields.

Correctional risk assessment applies it. Modern instruments, including the Level of Service Inventory-Revised, the Ohio Risk Assessment System, the federal Post Conviction Risk Assessment, and the proprietary COMPAS, score a person on factors that predict reoffending. Some are static and unchangeable, such as age at first arrest and prior convictions. Others are dynamic and treatable, such as substance use, employment, associates, and antisocial attitudes, which is what makes an instrument useful for case planning rather than merely for prediction.

How accurate are they? The standard metric is the area under the curve, or AUC, which has a clean interpretation: it is the probability that a randomly chosen person who does reoffend scored higher than a randomly chosen person who does not. An AUC of 0.5 is a coin flip. An AUC of 1.0 is perfect.

Well-validated correctional instruments generally achieve AUCs in the range of about 0.64 to 0.71. That is genuinely better than chance and genuinely far from certainty. Take a group of people classified as high risk: a majority of them will typically not be rearrested for a serious offence within the follow-up period. Any use of these tools that treats a high-risk classification as a prediction about an individual is misusing them.

What matters here: Risk instruments predict at the level of groups with moderate accuracy. An AUC near 0.70 means the ordering is informative and the individual prediction is not reliable, which is the distinction that most public argument about these tools ignores.

The fairness argument, and the result that resolves it

In May 2016 ProPublica published an analysis of COMPAS risk scores for 7,214 defendants in Broward County, Florida, matched to subsequent arrest records. Its central finding concerned errors. Among defendants who did not go on to be arrested again, Black defendants were roughly twice as likely as white defendants to have been classified as higher risk. Among those who did go on to reoffend, white defendants were more likely to have been labelled low risk.

Northpointe, the tool's developer, responded that the instrument was calibrated: within any given risk score, the observed reoffending rate was approximately the same for Black and white defendants. A score of seven meant roughly the same probability regardless of race, which is the property most people would name if asked what an unbiased score means.

So who was right? Both, and this is the most important thing in the lesson.

Jon Kleinberg, Sendhil Mullainathan, and Manish Raghavan, and independently Alexandra Chouldechova, proved a mathematical result. When the base rate of the outcome differs between two groups, and the instrument is not a perfect predictor, it is impossible for a risk score to satisfy all of the following at once: calibration within each group, equal false positive rates across groups, and equal false negative rates across groups. Only in degenerate cases, perfect prediction or identical base rates, can they all hold.

Read that again, because it is not a statement about COMPAS or about any particular developer. It is a theorem. In Broward County the rearrest base rates differed between the groups. Given that, a calibrated instrument must produce unequal error rates, and an instrument with equal error rates must be uncalibrated. Northpointe's defence and ProPublica's critique were both accurate descriptions of the same tool.

The consequence is uncomfortable and clarifying. There is no technical fix that satisfies every fairness definition simultaneously, so choosing among them is a normative decision about which kind of error to distribute how. That decision belongs to legislatures and to the public, not to a vendor's engineers, and pretending it is a technical matter is how it gets made by default.

Where risk assessment is used, and what it displaces

Risk scores now appear at pretrial release, in sentencing in some states, in prison classification, in programme assignment, in parole release, and in supervision intensity. They do different work in each.

In programme assignment they are least controversial, because the risk principle from the previous lesson makes matching intensity to risk the evidence-based approach.

In sentencing they are most controversial, because a sentence based on predicted future conduct sits awkwardly with the retributive principle from this course's first lesson, which holds that punishment tracks what was done rather than what is forecast. Research by Megan Stevenson and Jennifer Doleac on Virginia's sentencing risk assessment found that judges' use of the instrument was uneven and that its effects on sentences were considerably smaller than either its advocates or its critics expected, which is a reminder that a tool's effect depends on whether decision-makers actually follow it.

Common misconceptions

• Prisoners have a right to parole. Greenholtz held there is no constitutional entitlement to release before a valid sentence expires, though a mandatory state statute can create a liberty interest.
• Parole revocation requires the protections of a criminal trial. Morrissey requires notice, disclosure, a hearing before a neutral body, and written reasons, not a jury or proof beyond a reasonable doubt.
• The hungry judges study proves that parole decisions depend on meal breaks. The case ordering was not random, represented prisoners were scheduled earlier, and the effect size is implausible for the proposed mechanism.
• A high risk score predicts that this person will reoffend. At an AUC near 0.70, most people classified high risk are not rearrested for a serious offence in the follow-up period.
• A better algorithm could satisfy every fairness criterion. When group base rates differ and prediction is imperfect, calibration and equal error rates are mathematically incompatible.

Where this leaves us

• Morrissey v. Brewer requires a preliminary and a final revocation hearing with notice, disclosure, a neutral body, and written reasons, but not a jury or counsel as of right.
• Greenholtz held there is no constitutional right to parole, so the process protections attach to losing conditional liberty rather than to obtaining it.
• Releases shifted from predominantly discretionary parole in the 1970s to predominantly mandatory release, with discretionary parole still dominant in a substantial minority of states.
• The widely cited Israeli parole study documented a real pattern whose causal interpretation does not survive the finding that case ordering was not random.
• Actuarial instruments achieve AUCs of roughly 0.64 to 0.71, which is informative at the group level and unreliable for individual prediction.
• Calibration within groups and equal false positive and false negative rates cannot all hold when base rates differ, so choosing among fairness definitions is a normative decision rather than a technical one.

Sources

1. Morrissey v. Brewer, 408 U.S. 471 (1972). supreme.justia.com
2. Greenholtz v. Inmates of the Nebraska Penal and Correctional Complex, 442 U.S. 1 (1979). supreme.justia.com
3. Chouldechova, A. (2017). Fair prediction with disparate impact: A study of bias in recidivism prediction instruments. Big Data, 5(2), 153-163. doi.org
4. Danziger, S., Levav, J., and Avnaim-Pesso, L. (2011). Extraneous factors in judicial decisions. Proceedings of the National Academy of Sciences, 108(17), 6889-6892. doi.org
5. National Institute of Justice. (n.d.). Risk assessment and corrections research. Office of Justice Programs. nij.ojp.gov
6. Meehl, P. E. (1954). Clinical versus statistical prediction: A theoretical analysis and a review of the evidence. Minneapolis: University of Minnesota Press.

Forty thousand rules

More than 600,000 people leave American state and federal prisons every year, and several million more cycle out of local jails. The National Inventory of Collateral Consequences of Conviction, a catalogue compiled from federal and state statutes and regulations, lists more than 40,000 separate legal restrictions that attach to a criminal conviction in the United States.

They are called collateral consequences because they are not part of the sentence. No judge imposes them, no sentencing hearing considers them, and until fairly recently no lawyer was required to mention them. They attach automatically, by operation of law, and many of them last for life.

A person sentenced to eighteen months serves eighteen months and then encounters the other punishment, which nobody announced and which does not end.

The first seventy-two hours

Start with the practical, because the abstractions matter less than the sequence.

A person released from state prison typically receives a small sum of money, in many states a couple of hundred dollars or less, sometimes a bus ticket, and the clothes they were admitted in or a set issued for the purpose. They may or may not have identification. Many states release people without a state ID card, a birth certificate, or a Social Security card, which are the documents required to open a bank account, sign a lease, start a job, or obtain benefits. Getting them requires having the others.

They need somewhere to sleep tonight. If a parole plan named a relative's address, that address had to be approved, and it may have been rejected because someone else living there has a record. If it was not approved, the destination may be a shelter.

They need to report to a supervision officer within a set number of days, and the appointment will be during working hours, which conflicts with looking for work.

And, as the health lesson established, this is the period in which mortality runs roughly thirteen times the comparison rate, concentrated in overdose. The riskiest fortnight of a person's life begins on the day the state stops being responsible for them.

Key idea: Reentry failure is usually not a failure of resolve. It is a sequencing problem in which the documents, housing, income, and appointments a person needs each depend on already having one of the others.

The domains where a record operates

Domain	What the record does
Employment	Background checks, occupational licensing bars, statutory exclusions from whole industries
Housing	Federal public housing exclusions, private landlord screening, denial of subsidised units
Public benefits	Restrictions on food assistance and cash assistance for drug felonies, adopted federally in 1996 with state opt-outs
Voting	Disenfranchisement, permanent in some states and restored on release or on completion of supervision in others
Immigration	Deportability and inadmissibility, including for offences classified as aggravated felonies
Family	Termination of parental rights under child welfare timelines, and child support arrears accrued during confinement
Education	Disclosure requirements on applications, and until recently exclusion from federal student aid

Two of these deserve elaboration because they are structurally different from the others.

Occupational licensing. Roughly a quarter of American jobs require a state licence, and a large share of licensing statutes include a good moral character clause or an explicit bar for people with convictions. These often apply to occupations that prison vocational programmes actually train people for: barbering, food handling, nursing assistance, and various trades. It is entirely possible to complete a certified programme inside and be legally barred from the licence outside.

Voting. The Sentencing Project's periodic counts put the number of Americans disenfranchised because of a felony conviction at roughly 4.6 million in 2022, down from about 5.2 million two years earlier. The decline came from state reforms, most prominently Florida's Amendment 4 in 2018, though its implementation was subsequently conditioned on payment of outstanding fines and fees. Practice varies enormously: some states restore the vote automatically on release, some on completion of supervision, and a few impose permanent loss for certain offences.

One consequence was made constitutional. In Padilla v. Kentucky, decided in 2010, the Supreme Court held that defence counsel must advise a non-citizen client whether a plea carries a risk of deportation, and that failure to do so can constitute ineffective assistance. That decision treats one collateral consequence as sufficiently severe and sufficiently automatic that it cannot be left unmentioned. It has not been extended to the other thousands.

Ban the box, and a result nobody wanted

The most instructive policy episode in this area concerns a reform with obvious logic that produced an unintended effect.

Ban the box policies prohibit employers from asking about criminal history on an initial job application, deferring the question until later in the hiring process. The reasoning is that a checkbox screens people out before any human considers their qualifications, and that a person who reaches an interview can explain.

Amanda Agan and Sonja Starr tested it directly. They sent approximately 15,000 fictitious online job applications to employers in New Jersey and New York City, before and after ban-the-box policies took effect, randomising both applicant race and criminal record. Before the policy, employers who asked about records showed a modest gap in callbacks between white and Black applicants. After the policy removed the question, that racial gap widened substantially at those same employers.

The explanation is the one the discrimination literature predicts. Deprived of individual information about a record, some employers appear to have fallen back on group-level inference about who was likely to have one. Jennifer Doleac and Benjamin Hansen, using employment data rather than an audit design, reached a compatible conclusion: employment fell for young low-skilled Black and Hispanic men in jurisdictions adopting ban the box.

Now the honest qualification, because this result is frequently over-read. These studies measure aggregate effects at the callback and employment stage. They do not show that ban the box harms people who actually have records, and it may well help them by getting them further into a process where they can explain. The policy trade-off is real: a measure that helps the people it targets can, by removing information, harm a larger group through statistical inference.

What matters here: Removing information from a decision does not remove the decision. Where employers cannot observe an individual record, some substitute a guess based on group membership, which is why a well-intentioned reform produced a widened racial gap in callbacks.

What the reentry evaluations show

The evidence on reentry programming is more mixed than advocates suggest and more encouraging than the nothing-works tradition implies.

Transitional jobs have the strongest evaluation base. The Center for Employment Opportunities model provides immediate paid transitional work, paid daily, with supervision and job placement support. A large randomised evaluation found reductions in recidivism concentrated among participants who enrolled shortly after release, the highest-risk window, while the employment gains faded once the subsidised work ended. That pattern, real effects in the risk window and fading employment effects, recurs across the transitional employment literature.

Housing is the least evaluated and probably the most important. Analyses of the National Former Prisoner Survey found that formerly incarcerated people experience homelessness at something like ten times the rate of the general population, and the rate is far higher for people released multiple times. Programme evaluation here lags badly behind the size of the problem.

Record clearing has produced the most striking recent findings. J. J. Prescott and Sonja Starr studied Michigan's expungement system using state administrative data. Two results stand out. First, uptake is very low: only a small fraction of people legally eligible for expungement actually obtain it within years of becoming eligible, because the process requires knowing about it, filing, and often paying. Second, among those who do obtain it, subsequent crime rates are very low and wages rise substantially within two years. The combination, large benefits and tiny take-up, is the strongest possible argument for automatic clearing, and it is why clean slate statutes making expungement automatic on eligibility have spread across states.

Two structural points to carry

First, collateral consequences are legislative rather than judicial. They were enacted one at a time by legislatures and agencies, each addressing a specific concern, with no body assessing the cumulative effect. Nobody designed a system in which a single conviction triggers 40,000 possible restrictions; it accumulated.

Second, the consequences bite hardest on precisely the things that the desistance literature identifies as protective. Employment, housing, family ties, and civic participation are what the research says reduce reoffending. Those are the four domains the collateral consequence regime restricts most heavily. Whatever else can be said about it, the structure works against its own stated purpose.

Common misconceptions

• The sentence is the punishment. More than 40,000 legal restrictions attach automatically to a conviction, none of them imposed by a judge and many of them permanent.
• Vocational training inside solves the employment problem. Roughly a quarter of American jobs require a licence, and many licensing statutes bar people with convictions from precisely the trades prison programmes teach.
• Ban the box straightforwardly helps people with records. Audit evidence shows the racial callback gap widened when the question was removed, consistent with employers substituting group inference for individual information.
• People with records do not seek expungement because they do not want it. Michigan data show very low take-up among the eligible, driven by information, filing, and cost barriers, alongside large wage gains for those who obtain it.
• Reentry failure reflects a lack of motivation. Documents, housing, income, and appointments each depend on already having one of the others, which is a sequencing problem rather than a character problem.

The takeaway

• More than 600,000 people leave state and federal prisons each year, and over 40,000 legal collateral consequences attach automatically to a conviction.
• The first days after release combine missing identification, unapproved housing, no income, and mandatory appointments, in the period when mortality risk peaks.
• Collateral consequences operate across employment, housing, benefits, voting, immigration, family, and education, with occupational licensing and disenfranchisement the largest in scale.
• Padilla v. Kentucky made deportation advice a Sixth Amendment requirement, the one collateral consequence the Constitution now reaches.
• Agan and Starr found that ban-the-box policies widened the racial callback gap, consistent with employers substituting group inference when individual information is removed.
• Transitional jobs reduce recidivism in the immediate post-release window, and expungement produces large wage gains with very low take-up, which is the argument for automatic clearing.

Sources

1. Agan, A., and Starr, S. (2018). Ban the box, criminal records, and racial discrimination: A field experiment. Quarterly Journal of Economics, 133(1), 191-235. doi.org
2. Padilla v. Kentucky, 559 U.S. 356 (2010). supreme.justia.com
3. Council of State Governments Justice Center. (n.d.). National Inventory of Collateral Consequences of Conviction. csgjusticecenter.org
4. The Sentencing Project. (n.d.). Felony disenfranchisement research. sentencingproject.org
5. Prison Policy Initiative. (n.d.). Housing and employment after incarceration. prisonpolicy.org
6. Bureau of Justice Statistics. (n.d.). Reentry and recidivism statistics. Office of Justice Programs. bjs.ojp.gov

Every death sentence, void

On 29 June 1972 the Supreme Court decided Furman v. Georgia by five votes to four. The Court could not agree on an opinion, so it issued a brief unsigned judgement and then nine separate opinions, one from every justice. The holding was that the death penalty as then administered violated the Eighth and Fourteenth Amendments.

The reasoning that mattered was not that execution is inherently cruel; only two justices took that position. It was that the penalty was being imposed arbitrarily. Justice Potter Stewart's concurrence supplied the image that has framed the argument ever since: these death sentences were cruel and unusual in the same way that being struck by lightning is, freakishly and capriciously selected from among a much larger set of equally culpable defendants.

Roughly 600 people came off death rows across the country. Four years later the penalty returned, because states rewrote their statutes and the Court approved the rewrite.

This lesson puts both sides of the resulting argument as strongly as they can be put. That is not neutrality for its own sake. It is because the disagreement contains claims of two different kinds, and separating them is the only way to see what evidence could and could not settle.

The constitutional architecture

In Gregg v. Georgia, decided in 1976, the Court upheld statutes designed to answer Furman. The approved architecture has three components that still define capital procedure.

A bifurcated trial separates the determination of guilt from the determination of sentence, so that evidence about the defendant's life and character does not contaminate the guilt phase. Statutory aggravating factors narrow the class of murders eligible for death, and at least one must be found before a death sentence may be considered. Mitigating evidence must be admissible without limitation, and the sentencer must be able to give it effect. Automatic appellate review completes the structure.

Since then the Court has carved out categories of people who may not be executed at all, each time reasoning from evolving standards of decency measured largely by state legislative practice.

Case	Year	Holding
Furman v. Georgia	1972	Existing statutes unconstitutional as arbitrarily applied
Gregg v. Georgia	1976	Guided discretion statutes with bifurcated trials upheld
Atkins v. Virginia	2002	Execution of people with intellectual disability barred
Roper v. Simmons	2005	Execution for offences committed under 18 barred
Kennedy v. Louisiana	2008	Death barred for rape of a child where the victim did not die

The practice has contracted sharply. Executions peaked at 98 in 1999 and have run in the region of twenty a year recently. The death row population, near 3,600 around 2000, is now roughly 2,200. Around two dozen states have abolished capital punishment and several others operate executive moratoria, so the penalty is concentrated in a shrinking number of jurisdictions and, within them, in a small number of counties.

The case for, put properly

Five arguments, and the first is the one that carries the most weight among people who actually hold the position.

Desert. Some acts are so grave that no lesser penalty is proportionate to them. A person who tortures and murders a child has, on this view, forfeited a claim to continued life, and imposing a lesser penalty fails to take the wrong seriously. This is a retributive claim of the kind this course's first lesson analysed, and its crucial property is that it does not depend on any consequence. Show a retentionist that executions have no deterrent effect and you have not touched the argument, because it was never a deterrence argument.

Incapacitation. An executed person cannot kill again, including inside prison, where people already serving life have, on this view, nothing further to lose. The honest version acknowledges that life without parole achieves nearly all of this, and rests on the residual: homicides committed in custody and after escape are rare but not zero.

Expressive value. Law communicates. Reserving the ultimate penalty for the ultimate crime states, in the most emphatic terms available to a legal system, the value of the life that was taken. On this account abolition says something about how much a murdered person mattered.

Democratic authority. Capital punishment has commanded majority support in American polling for most of the modern era, peaking near 80 percent in the mid-1990s and running near or slightly above half in recent years. Where legislatures have enacted it and voters have endorsed it, its abolition by judicial decision is a countermajoritarian act requiring justification.

Deterrence. The claim that executions prevent future murders. This is the weakest of the five, and the next section explains why.

What the deterrence research shows

Beginning in the 1970s, econometric studies claimed to find that each execution prevented some number of murders. Other studies found nothing, and a few found the opposite. The dispute continued for three decades.

In 2012 the National Research Council convened a committee to assess the whole body of work. Its conclusion was unusually blunt for a consensus report. The committee found that the research to date is not informative about whether capital punishment decreases, increases, or has no effect on homicide rates, and it recommended that these studies not be used to inform deliberations about capital punishment.

The reasons are methodological and worth understanding because they generalise. The studies must model the sanction regime a potential murderer perceives, and no available data measure that. They must specify a counterfactual, meaning what the punishment would have been without the death penalty, and that alternative, usually life without parole, is itself severe and varies across states. Execution is rare and homicide is comparatively common, so the statistical leverage is very small. And results proved highly sensitive to model specification, with plausible alternative specifications producing effects of opposite sign.

What matters here: The authoritative scientific assessment is not that the death penalty fails to deter. It is that the available research cannot tell us either way, which means anyone citing deterrence in either direction is going beyond the evidence.

The case against, put properly

Innocence. Since 1973, close to 200 people have been exonerated and released from American death rows after their convictions were overturned. These are not cases where doubt was raised; they are cases where the conviction failed. Some involved DNA evidence, others recanted testimony or undisclosed exculpatory material. The argument is not that the system convicts innocent people at a higher rate in capital cases than elsewhere; it is that this is the one penalty in which an error cannot be corrected afterward.

Arbitrariness and race. David Baldus and colleagues analysed more than 2,000 Georgia murder cases, controlling for hundreds of variables. Their central finding was about the race of the victim rather than the defendant: defendants charged with killing white victims were roughly 4.3 times more likely to receive a death sentence than defendants charged with killing Black victims.

That study reached the Supreme Court in McCleskey v. Kemp, decided in 1987. The Court accepted the statistical validity of the study and rejected the claim anyway, holding that a defendant must show purposeful discrimination in his own case, and that a general statistical pattern is insufficient. Justice Lewis Powell, who wrote the majority opinion, said after his retirement that it was the vote he would most like to change. The case is also notable for what the majority said about the implications: if statistical disparities invalidated capital sentencing, similar arguments would call other sentencing decisions into question, which is a candid statement that the problem was too large to remedy rather than too small to matter.

Cost. Capital cases are far more expensive than cases resulting in life without parole. Bifurcated trials, qualified counsel requirements, extensive mitigation investigation, mandatory appellate review, and years of post-conviction litigation each add cost, and most death sentences are eventually overturned or commuted, so the expense is incurred for a penalty that is frequently not carried out. State-level cost studies have consistently found this.

Method. Lethal injection has generated its own litigation. In Baze v. Rees in 2008 and Glossip v. Gross in 2015, the Court upheld challenged protocols and, in Glossip, required a prisoner challenging a method to identify a known and available alternative that substantially reduces risk. Drug supply problems, as manufacturers refused sales for executions, have produced improvised protocols and a series of prolonged and evidently painful executions.

Concentration. A very small number of counties account for a large share of American death sentences. That concentration means the penalty depends heavily on the charging preferences of individual elected prosecutors, which is difficult to reconcile with the equal application Furman demanded.

Sorting the argument

Now the analytic work that this lesson exists to do.

Some of these claims are empirical and could in principle be settled. Whether executions deter is empirical, and the National Research Council's answer is that current methods cannot determine it. Whether capital cases cost more is empirical and the answer is yes. Whether innocent people are sentenced to death is empirical and the answer is demonstrably yes. Whether the race of the victim predicts a death sentence is empirical and the Baldus finding stands.

One claim is not empirical at all. Whether some crimes deserve death is a moral judgement, and no dataset speaks to it. A retentionist who holds the desert position can accept every empirical finding above and retain the position, arguing that the errors and disparities are reasons to improve the system rather than abandon the penalty. An abolitionist who holds that the state should never kill can likewise concede a deterrent effect if one were demonstrated and remain unmoved.

Where the two do meet is on administration. If the moral case requires that the penalty be reserved for the most culpable, then evidence that it is instead distributed by the victim's race, the county of prosecution, and the quality of appointed counsel is an argument the retributivist has to answer on their own terms. That is the point at which the empirical findings bear directly on the moral position, and it is where the serious argument now takes place.

Common misconceptions

• Furman held that the death penalty is inherently cruel. Only two justices took that view; the controlling reasoning was arbitrariness in application.
• Research has shown that the death penalty does not deter. The National Research Council concluded that the research cannot determine whether it decreases, increases, or has no effect.
• Executing people is cheaper than imprisoning them for life. Capital cases cost substantially more because of bifurcated trials, counsel requirements, mitigation investigation, and mandatory review.
• McCleskey rejected the Baldus study's statistics. The Court accepted the statistics and held that a general pattern cannot establish a constitutional violation without proof of purposeful discrimination in the individual case.
• The retributive argument is refuted by showing no deterrent effect. The desert claim never rested on consequences, so consequence evidence does not touch it directly; the administration evidence does.

Putting it together

• Furman voided every existing death sentence in 1972 on arbitrariness grounds, and Gregg restored the penalty in 1976 under bifurcated trials with statutory aggravators and unrestricted mitigation.
• Atkins, Roper, and Kennedy carved out categorical exclusions for intellectual disability, offences committed under 18, and child rape without a death.
• The strongest case for rests on desert, which makes no empirical prediction, supported by incapacitation, expressive, and democratic arguments.
• The National Research Council concluded in 2012 that deterrence research cannot determine the sign of the effect and should not be used to inform policy.
• The strongest case against rests on irreversible error, with close to 200 death row exonerations since 1973, on the Baldus finding of a 4.3-fold victim-race disparity, and on cost and geographic concentration.
• The empirical and moral claims are separable, and they meet on administration: evidence about how the penalty is actually distributed bears directly on a desert-based defence of it.

Sources

1. Furman v. Georgia, 408 U.S. 238 (1972). supreme.justia.com
2. Gregg v. Georgia, 428 U.S. 153 (1976). supreme.justia.com
3. McCleskey v. Kemp, 481 U.S. 279 (1987). supreme.justia.com
4. Roper v. Simmons, 543 U.S. 551 (2005). supreme.justia.com
5. Death Penalty Information Center. (n.d.). Facts and research on capital punishment. deathpenaltyinfo.org
6. National Research Council. (2012). Deterrence and the death penalty. Washington, DC: National Academies Press. nap.nationalacademies.org

From 105,000 to 25,000

In 1997 the federal Census of Juveniles in Residential Placement counted roughly 105,000 young people held in juvenile facilities in the United States on a single day. In the most recent counts the figure is around 25,000. That is a decline of more than three quarters, achieved over roughly two decades, during which juvenile arrests for violent offences also fell substantially.

It is the largest decarceration in American history, it happened almost without political controversy, and it happened in the system that had spent the 1990s being told it faced a coming wave of remorseless young criminals. That prediction, and what actually occurred, is where this lesson has to start.

Why there is a separate system at all

The first juvenile court in the world opened in Cook County, Illinois, in 1899. Its premise was parens patriae, the doctrine that the state may act as guardian for those who cannot protect themselves. A child who offended was not a criminal to be punished but a child in need of the state's care, and the court's job was to determine what the child needed rather than what the act deserved.

That premise produced a distinct vocabulary, which is not merely cosmetic. A young person is charged with a delinquent act rather than a crime. A hearing produces an adjudication rather than a conviction. The outcome is a disposition rather than a sentence. Proceedings are civil, records are confidential, and hearings have historically been closed.

Because the court was helping rather than punishing, the reformers reasoned, adversarial procedure was unnecessary and even counterproductive. There was no need for defence counsel, formal rules of evidence, or a standard of proof, because nobody was being prosecuted.

What informality cost: In re Gault

The flaw in that reasoning was demonstrated by one case.

In June 1964, Gerald Gault, aged fifteen, was taken into custody in Gila County, Arizona, after a neighbour complained of a lewd telephone call. His parents were not notified; his mother learned where he was from another family. There was no formal charge served, no lawyer, no witness against him at the hearing, no transcript, and no sworn testimony. The judge committed him to the State Industrial School until he reached twenty-one, a period of up to six years.

Had Gault been an adult, the maximum penalty for the same conduct under Arizona law was a fine of up to fifty dollars or two months in jail.

In In re Gault, decided in 1967, the Supreme Court held that juveniles facing proceedings that may result in commitment are entitled to written notice of the specific charges, the right to counsel including appointed counsel if indigent, the right to confront and cross-examine witnesses, and the privilege against self-incrimination. The opinion contained the sentence that broke the parens patriae rationale open: under our Constitution, the condition of being a boy does not justify a kangaroo court.

In re Winship added the reasonable doubt standard in 1970. But in McKeiver v. Pennsylvania in 1971 the Court declined to require jury trials in juvenile proceedings, reasoning that a jury would formalise the process to a degree inconsistent with what remained of its rehabilitative character. So the juvenile court today is a hybrid: most of the procedural protections of a criminal trial, without a jury, in a civil proceeding whose consequences can be severe.

Key idea: The juvenile court was built on the premise that it was helping rather than punishing, which was used to justify dispensing with procedure. Gault established that the consequences were severe enough to require the procedure regardless of the label.

The prediction that did not happen

In 1995 the political scientist John DiIulio published an essay warning of a coming generation of juvenile superpredators: young offenders he characterised as radically impulsive and remorseless, whose numbers he projected would rise sharply through the following decade. The term entered political speech immediately, and between 1992 and 1997 nearly every American state made it easier to prosecute young people as adults.

The predicted wave did not arrive. Juvenile violent crime arrest rates peaked in the mid-1990s and then fell steeply and continuously. DiIulio subsequently repudiated the forecast, and later joined a brief to the Supreme Court in a juvenile sentencing case arguing against the policies his own prediction had helped produce.

Two lessons are worth extracting. Criminological forecasts based on demographic projections alone have a poor record, because they hold offending rates per capita constant and those rates move. And a policy enacted in response to a forecast usually outlives the forecast, because repeal requires the same political energy that enactment did and there is no crisis to supply it.

Transfer to adult court, and what it produces

Young people reach adult court by three routes. Judicial waiver is a juvenile court judge's decision after a hearing. Prosecutorial direct file lets a prosecutor choose the forum without any judicial finding. Statutory exclusion removes categories of offence or age from juvenile jurisdiction entirely, so no discretion is exercised at all.

The purpose was deterrence and incapacitation, and the effect has been measured. A systematic review conducted for the Community Preventive Services Task Force, published through the Centers for Disease Control and Prevention, examined studies comparing young people transferred to adult court with comparable young people retained in the juvenile system. The finding was that transfer was associated with higher rates of subsequent violence, not lower. The task force recommended against the use of transfer for the purpose of reducing violence.

The mechanisms are the ones this course has already established. Adult facilities offer less age-appropriate programming and education. A young person in an adult prison is placed among older offenders, supplying exactly the antisocial associates the criminogenic need literature identifies. An adult conviction carries collateral consequences a juvenile adjudication does not. And a young person in an adult facility faces elevated risks of victimisation and of suicide.

What the developmental science added

The Supreme Court's juvenile sentencing decisions since 2005 rest heavily on developmental research, and it is worth knowing what that research actually says rather than the slogan version.

Laurence Steinberg and colleagues describe a dual systems pattern. Sensation seeking and reward sensitivity rise sharply in early adolescence and peak in the mid-teens. Capacities for impulse control, future orientation, and resistance to peer influence mature more slowly, continuing into the early twenties. The gap between the two curves is widest in mid-adolescence, which is precisely when risk-taking behaviour peaks.

Two features of that finding matter legally. It implies reduced culpability, since the capacity for self-regulation is not fully developed. And it implies greater capacity for change, since the person is still developing, which undercuts any judgement of permanent incorrigibility.

The Court used both. Roper v. Simmons barred capital punishment for offences committed under 18 in 2005. Graham v. Florida barred life without parole for non-homicide offences committed by juveniles in 2010. Miller v. Alabama held mandatory life without parole for juveniles unconstitutional in 2012, requiring individualised consideration of youth, and Montgomery v. Louisiana made that retroactive in 2016. The Court subsequently held, in 2021, that a sentencer imposing such a sentence need not make a separate factual finding of permanent incorrigibility, which narrowed the practical effect of the earlier decisions considerably.

Terrie Moffitt's developmental taxonomy supplies the criminological complement. She distinguished a large group whose antisocial behaviour is adolescence-limited, beginning and ending with adolescence, from a much smaller group whose antisocial behaviour is life-course-persistent, beginning early and continuing. Most delinquency belongs to the first group, and most of it stops without any intervention at all, which is a fact any evaluation of juvenile programming has to contend with.

What works, and what harms

The juvenile field has a stronger evidence base than adult corrections, because more of its interventions have been tested with randomised designs.

Family-based interventions have the best evidence. Multisystemic Therapy works intensively with the young person's whole environment, family, school, and peers, with therapists carrying small caseloads and being available around the clock. Functional Family Therapy targets family communication and problem solving over a shorter course. Treatment foster care places a young person with trained foster parents rather than in a group setting. All three have multiple randomised trials showing reductions in reoffending and out-of-home placement.

Notice what they share. All are delivered in the community rather than in an institution, all treat the family and peer environment rather than the individual alone, and all avoid grouping delinquent young people together.

That last point explains the failures. Scared Straight increases offending, as the programming lesson established. Juvenile boot camps show no reliable effect. Large congregate institutions produce poor outcomes and, repeatedly across states, documented abuse. A plausible common mechanism is deviancy training: putting delinquent adolescents together in an unstructured group lets them reinforce each other's behaviour, which is the antisocial associates need being actively supplied by the intervention.

The institutional alternative that performs best is the Missouri model, which replaced large training schools with small regional facilities holding roughly ten to twelve young people, staffed by youth specialists rather than guards, organised around continuous group treatment and education, in settings that deliberately do not look like prisons. Missouri reports low recidivism, and several states have adapted the approach. The evidence is observational rather than experimental, and it should be described that way.

Common misconceptions

• Juvenile court is criminal court for children. It is a civil proceeding using different vocabulary, without a jury, though Gault imported most of the procedural protections of a criminal trial.
• The superpredator forecast was borne out. Juvenile violent crime arrests fell steeply after the mid-1990s, and its author repudiated the prediction.
• Trying young people as adults reduces violence. The systematic review conducted for the Community Preventive Services Task Force found transfer associated with higher subsequent violence and recommended against it for that purpose.
• Adolescent brain research says teenagers cannot control themselves. It describes a gap between early-peaking sensation seeking and later-maturing self-regulation, which supports both reduced culpability and greater capacity for change.
• Grouping delinquent youth for treatment is efficient. Deviancy training in unstructured groups can increase offending, which is why the best-evidenced programmes work with families in the community instead.

What to carry forward

• The juvenile residential population fell from roughly 105,000 in 1997 to around 25,000, the largest American decarceration on record.
• The juvenile court rests on parens patriae, with delinquent acts, adjudications, and dispositions replacing crimes, convictions, and sentences.
• In re Gault required notice, counsel, confrontation, and the privilege against self-incrimination after a fifteen-year-old was committed for up to six years for conduct carrying a fifty dollar adult fine; McKeiver withheld the jury right.
• The superpredator forecast produced transfer laws in nearly every state and was then repudiated by its author as juvenile violence fell.
• Transfer to adult court is associated with higher subsequent violence, and the Community Preventive Services Task Force recommended against it for violence prevention.
• Multisystemic Therapy, Functional Family Therapy, and treatment foster care have randomised evidence of effectiveness; Scared Straight, boot camps, and large congregate institutions do not, plausibly because of deviancy training.

Sources

1. Office of Juvenile Justice and Delinquency Prevention. (n.d.). Statistics and research on juveniles in residential placement. Office of Justice Programs. ojjdp.ojp.gov
2. In re Gault, 387 U.S. 1 (1967). supreme.justia.com
3. Graham v. Florida, 560 U.S. 48 (2010). supreme.justia.com
4. Miller v. Alabama, 567 U.S. 460 (2012). supreme.justia.com
5. National Institute of Justice. (n.d.). CrimeSolutions: Juvenile programme and practice ratings. Office of Justice Programs. crimesolutions.ojp.gov
6. Moffitt, T. E. (1993). Adolescence-limited and life-course-persistent antisocial behavior: A developmental taxonomy. Psychological Review, 100(4), 674-701.

Hamilton County, 1984

The Corrections Corporation of America was incorporated in Nashville in 1983. The following year it took over operation of the county corrections facility in Hamilton County, Tennessee, becoming the first company in American history to run a secure adult correctional institution under contract to a government. Within a few years a second firm, Wackenhut, was competing for the same business. Those two companies, now trading as CoreCivic and the GEO Group, still dominate the industry.

Today roughly 8 percent of people in American state and federal prisons are held in privately operated facilities, on the order of 90,000 to 100,000 people. That figure is small and it is also misleading in two directions.

It understates the concentration. Private incarceration is not spread evenly. Several states use none at all. A handful contract out a fifth or more of their prison population, so the debate is a live one in those states and abstract in others.

And it excludes immigration detention, where the large majority of capacity is privately operated. Immigration detention is civil rather than criminal, which is why it does not appear in prison counts, and it is the sector where the industry's growth has been concentrated.

The case for contracting, stated fairly

The argument is not simply that companies are efficient. It has four parts, and each has some force.

Cost discipline. A firm that must bid against competitors has a reason to find savings that a public agency with an appropriated budget does not. Public corrections budgets grow through legislative processes that reward avoiding conflict with employee unions.

Capacity flexibility. Building a prison requires a capital appropriation and years of construction. A contract can add beds within months and release them when demand falls, which matters to states facing court-ordered population caps.

Contractual specification. A contract can require performance standards, accreditation, staffing ratios, and programme delivery, with financial penalties for failure. A civil service system cannot easily impose comparable consequences on itself.

Benchmarking. A state with both public and contracted facilities has a comparator. A state with only public facilities has no external standard against which to measure cost or performance.

The argument that actually matters: incomplete contracts

The strongest analytical objection is not about greed. It comes from contract theory, and prisons are its textbook illustration.

Oliver Hart, Andrei Shleifer, and Robert Vishny set out the framework in a 1997 paper on the proper scope of government, using prisons as their central example. Their question is when a government should provide a service directly and when it should buy it.

Their answer turns on what can be written into a contract. A private provider has strong incentives both to reduce cost and to improve quality. Where quality is easy to specify and verify, that combination is good: the buyer writes the standard, monitors it, and enforces it. Where important dimensions of quality cannot be specified and verified, the cost-reduction incentive dominates, because savings are captured immediately and the quality loss is not contractible.

Now apply that to a prison. Some things are easy to write down: bed count, calorie provision, hours of programming offered, staffing ratios on paper. Other things are not. Whether officers treat people with a minimum of dignity. Whether force is used when it is not necessary. Whether a request for medical attention is taken seriously at three in the morning. Whether the atmosphere on a tier is one in which people can be safe.

Those unspecifiable dimensions are exactly what determines whether a prison is decent. And the cheapest way to run a prison is to hire fewer officers, pay them less, train them less, and accept the higher turnover that follows, all of which degrade precisely the qualities the contract cannot measure.

Key idea: The incomplete contracts objection does not assume bad faith. It says that where quality is unverifiable, the incentive to cut cost dominates by construction, and that prisons have an unusually large share of unverifiable quality.

What the evidence shows on cost

Less than either side claims, and the reason is instructive.

The Government Accountability Office reviewed the comparative cost studies and concluded that they did not permit firm conclusions about whether private operation saves money. Academic reviews since have generally agreed. The problem is not that nobody has looked; it is that the comparison is very hard to make honestly, for three reasons.

Population differences. Contracts routinely exclude the most expensive prisoners: those requiring maximum security, those with serious chronic medical conditions, those with major psychiatric needs. A facility holding a younger, healthier, lower-security population will cost less per person regardless of who operates it. This is called cream skimming, and any comparison that does not adjust for it is measuring the selection rather than the operator.

Accounting boundaries. Public per-prisoner costs frequently exclude items that appear elsewhere in a state budget: pension liabilities, central administration, capital costs, legal defence, and the cost of monitoring the contracts themselves. Including or excluding these changes the answer.

Contract terms. Some contracts have included occupancy guarantees, sometimes called lockup quotas, requiring the state to pay for a minimum number of beds whether or not they are used. Under such a term, a state that successfully reduces its prison population pays for empty cells, which changes what the comparison is even about.

What the evidence shows on conditions

Here there is a more definite finding, from the government's own auditor.

In 2016 the Office of the Inspector General of the Department of Justice compared federal contract prisons with comparable Bureau of Prisons institutions across a set of safety and security measures. Contract prisons had more incidents per capita in most of the categories examined, including contraband, assaults, uses of force, and lockdowns. The report also identified specific operational failures, including one facility where new arrivals were housed in solitary confinement units because no other space was available.

The policy consequences oscillated with administrations. A 2016 memorandum from the Deputy Attorney General directed the Bureau of Prisons to phase out its use of contract facilities. It was rescinded in early 2017. An executive order in January 2021 directed the Department of Justice not to renew contracts with privately operated criminal detention facilities. That order reaches federal criminal detention only; it does not cover immigration detention, which is administered by a different department, and it does not reach state prisons at all.

One further study is worth knowing because it identifies a mechanism nobody had looked for. Anita Mukherjee, analysing Mississippi data, found that people held in private facilities accumulated more conduct violations than comparable people in public facilities, and consequently served longer before release. Conduct violations reduce good-time credits. If a facility is paid per occupied bed, the incentive runs toward writing more violations. It is one study in one state, and it is the clearest illustration available of how a payment structure can reach into the length of a sentence.

The privatisation that happens inside public prisons

Focusing on who owns the building misses most of the money.

Publicly operated prisons routinely contract out food service, medical and mental health care, commissary sales, transportation, electronic monitoring, money transfer services, and telephone and video communication. Each of these is a market with its own incentives, and several are structured so that the people paying are prisoners and their families rather than the state.

Prison telephone rates became the clearest example. For years, correctional telephone contracts were awarded partly on the basis of commissions paid back to the agency from call revenue, which gave agencies an incentive to select the highest-priced bidder. Families, disproportionately poor, paid rates far above ordinary consumer rates for calls that the desistance literature identifies as protective, since maintained family contact is associated with better post-release outcomes. The Federal Communications Commission regulated interstate rates and, after the Martha Wright-Reed legislation enacted in early 2023, obtained authority over in-state calls as well.

The general point matters more than the example. A prison can be publicly owned and publicly staffed while most of the services within it are sold, sometimes to a captive market with no alternative supplier.

Common misconceptions

• Private prisons drive American mass incarceration. They hold roughly 8 percent of state and federal prisoners, so the scale of incarceration is overwhelmingly a public phenomenon.
• Private operation has been shown to save money. The Government Accountability Office and subsequent reviews found the comparisons inconclusive, largely because of cream skimming and accounting boundaries.
• The objection to contracting is that companies are greedy. The incomplete contracts argument assumes ordinary incentives and turns on which dimensions of quality can be written into a contract and verified.
• The 2021 executive order ended private incarceration. It applies to federal criminal detention, not to immigration detention and not to state prisons, where most private capacity sits.
• Privatisation means the whole prison is private. Food, health care, commissary, transport, and communications are contracted out in most public prisons, often to markets in which prisoners and families are the paying customers.

What you now know

• Private corrections began with the Corrections Corporation of America's 1984 contract in Hamilton County, Tennessee, and is now dominated by two firms.
• Roughly 8 percent of state and federal prisoners are in private facilities, concentrated in a minority of states, while the large majority of immigration detention capacity is privately operated.
• The incomplete contracts framework predicts that where quality cannot be specified and verified, the cost-reduction incentive dominates, and prisons are the canonical case.
• Comparative cost studies are inconclusive because contracts exclude expensive prisoners and public accounting boundaries vary.
• The Justice Department Inspector General found more safety and security incidents per capita in federal contract prisons than in comparable Bureau of Prisons institutions.
• Services inside publicly run prisons, including health care, commissary, and telephone calls, are widely contracted out to markets in which prisoners and their families are the paying customers.

Sources

1. Bureau of Justice Statistics. (n.d.). Prisoners in privately operated facilities. Office of Justice Programs. bjs.ojp.gov
2. Office of the Inspector General, U.S. Department of Justice. (2016). Review of the Federal Bureau of Prisons' monitoring of contract prisons. oig.justice.gov
3. Hart, O., Shleifer, A., and Vishny, R. W. (1997). The proper scope of government: Theory and an application to prisons. Quarterly Journal of Economics, 112(4), 1127-1161. doi.org
4. U.S. Government Accountability Office. (n.d.). Reports on private correctional facilities and costs. gao.gov
5. Federal Communications Commission. (n.d.). Incarcerated people's communications services. fcc.gov

Two hundred thousand, then 1.6 million

In 1972 American state and federal prisons held about 200,000 people. The imprisonment rate was close to 100 per 100,000 residents, and it had been remarkably stable at roughly that level for the previous fifty years. Criminologists had a name for the stability, the theory of the homeostasis of punishment, and they treated it as one of the few durable regularities their field had produced.

Then it broke. The population rose every single year for the next thirty-seven years, reaching about 1.61 million in 2009, with an imprisonment rate near 500 per 100,000. Adding jails, the total confined population passed 2.2 million.

Since 2009 it has fallen, to roughly 1.2 million in state and federal prisons, with a sharp drop during the pandemic and a partial rebound afterward. That is a decline of about a quarter from the peak, which is real and which leaves the United States still incarcerating at several times the rate of any comparable democracy.

This lesson asks two questions. What produced the growth? And what does the evidence say about reversing it?

It was not crime

The intuitive explanation is that crime rose and imprisonment followed. The timing rules it out.

American crime rose through the 1960s and 1970s, was roughly flat through the 1980s with a violent spike around 1990, and then fell continuously and steeply from 1991 through the 2010s. Imprisonment rose throughout all of that, in periods of rising crime and falling crime alike. A variable that moves in one direction for thirty-seven years cannot be explained by a variable that moves up and then down.

The National Research Council's 2014 consensus report, the most authoritative assessment available, reached exactly that conclusion. The growth in American incarceration was driven by policy changes rather than by changes in offending: more people were sent to prison per arrest, and those sent stayed longer.

Decomposing which of the two mattered more is genuinely difficult, and the answer changed over time. In the earlier decades, rising admissions did more of the work. In the later decades, admissions flattened while time served kept rising as truth in sentencing, mandatory minimums, habitual offender laws, and the abolition of discretionary parole took effect. That distinction matters for policy, because reducing admissions and reducing time served require entirely different legislative changes.

Key idea: Prison growth ran continuously through periods of rising and falling crime, so crime cannot explain it. The National Research Council attributed it to policy: more admissions per arrest and, increasingly, longer time served.

The standard story, and where it fails

The account most people carry has three elements: the war on drugs, private prisons, and mandatory minimums. Each contains something true and none is sufficient.

The drug war. Drug enforcement absolutely drove enormous numbers of arrests and it transformed the federal system, where drug offences account for close to half the population. But the federal system holds about an eighth of American prisoners. In state prisons, which hold the other seven eighths, drug offences account for roughly one in seven people. Releasing everyone imprisoned for a drug offence nationally would reduce the total prison population meaningfully and would not come close to ending mass incarceration.

Private prisons. They hold about 8 percent of prisoners. Whatever their other problems, an industry of that size did not produce a fivefold increase in a public system.

Mandatory minimums. Real and consequential, particularly federally. They do not explain state systems where most sentences are not subject to them.

The composition fact that reorganises the whole discussion is this: more than half of state prisoners are serving sentences for offences classified as violent. Any policy programme that excludes violent offences from consideration cannot substantially reduce American incarceration. That is not an argument for or against any particular reform. It is arithmetic, and it is the single most consequential thing in this lesson.

Pfaff's argument: look at the prosecutor

John Pfaff proposed a different account, and it deserves close attention because it points at an actor nobody had been watching.

His observation concerns the period from roughly 1994 to 2008, when the prison population was still rising. During that period, arrests were flat or falling, in line with falling crime. But the number of felony cases filed in state courts rose substantially. On his reading of the state court data, the number of felony filings per arrest roughly doubled.

If arrests are flat and prison admissions rise, something between arrest and prison changed, and the decision that sits there is the prosecutor's charging decision. Pfaff also documented substantial growth in the number of line prosecutors during a period of falling crime, and he emphasised that prosecutors are elected locally, operate with almost no data collection about their decisions, and face essentially no external review.

The argument has been contested. Critics note that state court filing statistics are assembled from jurisdictions with inconsistent definitions and coverage, that the trend is sensitive to which states are included, and that charging decisions respond to the sentencing statutes legislatures pass, so prosecutorial discretion is exercised within a menu that others wrote.

What is not seriously contested is the structural point. Sentencing reforms of the 1980s constrained judges and abolished parole boards while leaving prosecutorial charging untouched and unmeasured. The system's least visible discretionary actor became its most powerful one, and we still have very little data about how that power is used.

Racial concentration

The growth was not distributed evenly. Black Americans are imprisoned at roughly five times the white rate, and this ratio, while still very large, has narrowed considerably from something closer to eight to one around 2000.

The most cited figure in this area needs careful handling. A Bureau of Justice Statistics projection published in 2003 estimated that, at the imprisonment rates then prevailing, about 32 percent of Black males born in 2001 would go to state or federal prison at some point in their lives. That was a lifetime projection built on 2001 rates. Imprisonment rates have fallen substantially since, and the projection therefore overstates current lifetime risk. It remains an accurate description of what the system was doing at its peak, and it is frequently quoted as though it described today.

James Forman's account complicates the political history in a way worth knowing. Studying the District of Columbia, where Black officials held the mayoralty, the council, the police leadership, and the prosecutor's office, he documented substantial Black community support for tougher drug and gun enforcement in the 1970s and 1980s, arising from the experience of neighbourhoods facing serious violence and receiving inadequate protection. The punitive turn was not simply imposed from outside on communities that uniformly opposed it, and any account that says so is missing what the record shows.

What decarceration has actually produced

The United States has now run several large natural experiments, and their results are more informative than any amount of argument.

California realignment. Faced with the population cap upheld in Brown v. Plata, California enacted Assembly Bill 109 in 2011, shifting responsibility for people convicted of lower-level felonies from state prisons to counties. The state prison population fell by tens of thousands within a year. Magnus Lofstrom and Steven Raphael examined the crime consequences using the variation in how counties were affected. They found no detectable increase in violent crime and a measurable increase in property crime concentrated in motor vehicle theft. They also computed the implied trade-off, and the number of additional property crimes per prison year avoided was small relative to the cost of the prison year.

New York City. The city's jail population fell from roughly 22,000 in the early 1990s to a small fraction of that over three decades, while the city recorded one of the largest crime declines of any American city. Two things fell together over the same period, which is not proof of causation in either direction and is powerful evidence against the claim that decarceration requires accepting more crime.

The First Step Act. The 2018 federal statute expanded earned time credits, made the Fair Sentencing Act's crack cocaine changes retroactive, reformed the stacking of firearm counts, and broadened compassionate release. Federal reporting on people released under its provisions has found recidivism rates comparable to or lower than those of other released federal prisoners, which is what the age-crime curve would predict for a group whose releases were often advanced by years.

What the honest version concedes. The realignment result is not zero. Property crime rose measurably, and the people victimised are real. Incapacitation is real, and reducing a prison population by any means will prevent fewer offences than not reducing it. The argument for decarceration is not that its crime cost is nil; it is that at American scale the crime cost per prison year is small, and that the same money spent differently buys more crime reduction. That is a comparative claim, and it is the one a serious critic should engage.

Where the field agrees

Correctional research disagrees about a great deal. Four things command broad agreement, and they are a reasonable place to end a course.

The growth was policy-driven rather than crime-driven. The marginal crime-prevention return to incarceration at American scale is low, because the marginal prisoner offends at a much lower rate than the average one. Supervision violations account for a large and reducible share of prison admissions, and reducing them requires no change to sentencing law at all. And no substantial reduction is possible without addressing sentence lengths for serious offences, because that is where most of the people are.

The disagreements that remain are about how much risk to accept, whom to ask to bear it, and what punishment is for. Those are the questions this course opened with, and the evidence you now have does not settle them. It constrains them, which is the most a course like this can honestly do.

Common misconceptions

• Prison growth tracked crime rates. The population rose every year from 1972 to 2009 through both rising and falling crime, which is why the National Research Council attributed it to policy.
• Ending the war on drugs would end mass incarceration. Drug offences account for roughly one in seven state prisoners, and state systems hold seven eighths of the total.
• One in three Black men will go to prison. That was a lifetime projection at 2001 rates; imprisonment rates have fallen substantially since, so it overstates current risk.
• The punitive turn was imposed entirely from outside affected communities. Forman's District of Columbia study documents substantial Black official and community support for tougher enforcement in the 1970s and 1980s.
• Decarceration has no crime cost. California realignment produced a measurable increase in motor vehicle theft; the argument is that the cost per prison year avoided is small, not that it is zero.

Where this leaves us

• The state and federal prison population went from about 200,000 in 1972 to roughly 1.61 million in 2009, and has fallen by about a quarter since.
• Crime cannot explain the growth, which ran continuously through rising and falling crime; the National Research Council attributed it to more admissions per arrest and longer time served.
• The drug war, private prisons, and mandatory minimums each explain part of the growth and none is sufficient, while more than half of state prisoners are held for offences classified as violent.
• Pfaff's account points at prosecutorial charging, noting that felony filings per arrest roughly doubled while arrests were flat, and the structural point about unmeasured prosecutorial discretion is not contested.
• Black Americans are imprisoned at roughly five times the white rate, down from about eight to one around 2000, and the widely quoted lifetime projection describes 2001 rates rather than current ones.
• California realignment produced no detectable rise in violent crime and a measurable rise in vehicle theft, and New York City's jail population fell alongside one of the largest crime declines in the country.

Sources

1. National Research Council. (2014). The growth of incarceration in the United States: Exploring causes and consequences. Washington, DC: National Academies Press. nap.nationalacademies.org
2. Lofstrom, M., and Raphael, S. (2016). Incarceration and crime: Evidence from California's public safety realignment reform. The ANNALS of the American Academy of Political and Social Science, 664(1), 196-220. doi.org
3. Bureau of Justice Statistics. (n.d.). Prisoners series and correctional populations. Office of Justice Programs. bjs.ojp.gov
4. United States Sentencing Commission. (n.d.). First Step Act implementation and recidivism reports. ussc.gov
5. Vera Institute of Justice. (n.d.). Incarceration trends. vera.org
6. Pfaff, J. F. (2017). Locked in: The true causes of mass incarceration and how to achieve real reform. New York: Basic Books.