Who Gets Punished, and Why

The labelling tradition asked who gets labelled. The measurement lesson showed that the crime statistics are the output of a five-gate machine. This lesson puts the two together and asks the question they lead to: not who offends, but who is punished.

And the answer is that the two populations — those who offend and those who are imprisoned — are not the same population, and the difference between them is patterned by exactly the inequalities of Part 8. Disentangling how much of that gap is differential offending and how much is differential treatment is one of the hardest and most important problems in the field, and this lesson is about doing it honestly.

Two drug users, one law, two lives.

Take two people who use the same illegal drug at the same rate.

The first is a university student who buys from a friend, uses at home and at private parties, and lives in a neighbourhood the police do not patrol for drugs. In the years he uses, he is never stopped, never searched, never arrested. He finishes his degree, enters a profession, and his drug use leaves no trace on his life. It is, in the terms of 10.1.3, primary deviance that was never labelled.

The second buys from a street dealer, because that is the market available to him, in a neighbourhood that is heavily policed for drugs. He is stopped and searched — one of many stops, most of which find nothing — and on one of them, something is found. He is arrested, charged, and acquires a record. The record closes off employment (see 7.4.2). He is stopped again, because he is now known. His drug use, identical in quantity to the student's, becomes the organising fact of his life.

Now the crucial point, and it has to be stated with care. Self-report surveys — where people anonymously report their own drug use (see 7.4.1) — consistently find that drug use rates are broadly similar across social and racial groups, and in some studies higher among the more advantaged. But arrests, charges and imprisonment for drug offences are dramatically concentrated among the poor and among minority groups.

Read the two facts together. If use is similar across groups but punishment is wildly unequal, then the inequality in punishment is not produced by inequality in offending. It is produced somewhere in the five gates — in who is policed, who is stopped, who is searched, who is charged, who is convicted, and who is sentenced to what.

This is the cleanest available case, because drug use is measurable by self-report and is roughly equally distributed. For most crimes it is far harder, because offending is genuinely unequal — and separating the differential offending from the differential treatment is the whole analytical problem. But the drug case establishes the principle: the gap between who offends and who is punished can be produced entirely by the response, and sometimes demonstrably is.

The disparity is not in dispute. Its cause is, and the causes are separable.

Everywhere it has been measured, the punished population is drawn disproportionately from the poor, the young, men, and — in societies with racialised hierarchies — from subordinated racial and ethnic groups (see 8.4). This is not contested; the prison populations of the world are among the most socially skewed populations there are.

What is contested is why, and there are two families of explanation, and the honest answer involves both.

Differential offending: the disparity reflects real differences in the rate and seriousness of offending, which themselves flow from the structural conditions of Part 8 — concentrated disadvantage, weak labour markets, the strain and opportunity structures of 10.1.2, the neighbourhood effects of 8.3.2. On this account the criminal justice system is processing a real difference in behaviour that inequality produced upstream.

Differential treatment: the disparity reflects unequal processing at the gates — more policing of some areas and groups, more stops, more charges, harsher sentences — for the same behaviour. On this account the system adds inequality to whatever the behaviour contributed.

These are not mutually exclusive, and the evidence is that both operate, in proportions that differ by offence and by stage. The task is to estimate each without letting a prior commitment decide the answer — which is 7.6.2's confounding problem in its most politically charged form, and where the field's methodological discipline matters most.

Disentangling the two, stage by stage

The analytical problem, and why it is so hard.

You cannot simply compare punishment rates across groups and call the gap discrimination, because the groups differ in offending, and offending is confounded with everything (see 7.6.2). Nor can you control for "offending" and call the residual discrimination, because the measure of offending is itself the arrest and conviction data produced by the system you are trying to evaluate — which is circular (see 8.4.2 on the same bad-control problem for hiring). If the police arrest one group more for the same behaviour, that group's recorded offending is higher, and controlling for it removes the discrimination from the estimate.

So the field's best evidence comes from designs that break the circle, and each stage needs its own.

Policing and stops. Where stop-and-search data can be compared against a benchmark of who is actually carrying contraband, the test is the hit rate: if one group is stopped far more but contraband is found on them at a lower rate, that is evidence the threshold for stopping them is lower — i.e. discrimination (an "outcome test"). Studies applying this logic have repeatedly found lower hit rates for more-stopped groups, consistent with differential treatment at the stop stage. The method has limitations (the benchmark problem is real), but it is far better than a raw comparison.

Charging and prosecution. Prosecutorial discretion is enormous and hard to study because it is opaque. Studies with good data find disparities in charging decisions and in plea bargaining that survive controls for offence and record — for example, differences in who is offered a reduced charge — though the evidence is patchier than for stops.

Sentencing. The most studied stage, because sentencing data is rich. The careful finding: after controlling for offence seriousness and prior record, sentencing disparities by race and class shrink substantially but do not vanish, with residual disparities concentrated in discretionary decisions and in the treatment of prior record. A crucial subtlety: prior record is itself a product of earlier differential treatment, so controlling for it — standard practice — removes part of the cumulative disadvantage from the estimate (see 8.4.2). The stage-by-stage disparities are individually modest and cumulatively large, because they compound: a small edge at each of five gates multiplies.

And the honest synthesis: at each stage, some of the disparity is differential offending and some is differential treatment; the treatment component is real, is largest where discretion is greatest, and — because it compounds across stages and because prior record launders earlier disparities into later "justified" ones — the cumulative differential treatment is substantially larger than any single stage reveals.

Mass incarceration

The scale, and what caused it — which is not what caused crime.

The United States is the extreme case, and it is worth stating plainly because the scale is genuinely exceptional. From the early 1970s, its imprisoned population grew roughly fivefold, reaching a level far above any comparable country and above its own historical range — an increase with few parallels in the history of democracies.

The decisive finding: this was not driven by crime. Crime rose and then fell (see 10.2.2); imprisonment rose throughout, including through the two decades when crime was falling. The two series diverge, which means incarceration was not tracking offending.

What drove it was policy — a set of choices, not a natural response to crime: harsher sentencing (mandatory minimums, three-strikes laws, truth-in-sentencing), the war on drugs (which filled prisons with drug offenders — see the story), the expansion of what was punished with prison, and longer sentences for the same offences. Each was a decision, made politically, and reversible (see 9.7.1 — this is an institution built by choice, and the countries that did not make these choices did not build these prisons).

And the consequences concentrate. Because the punished population is socially skewed, mass incarceration fell overwhelmingly on poor and minority communities — to the point where, for some groups in some places, imprisonment became a modal life event, more common than completing college. Which turns it from an individual outcome into a collective condition: a feature of whole neighbourhoods, reshaping families, labour markets, and the next generation's chances (see 8.3.3 — a father's incarceration is one of 8.3.3's channels, operating at community scale).

This is the point where the sociology of punishment rejoins the sociology of inequality. Mass incarceration is not only a response to disadvantage; it is a producer of it — removing people from labour markets, saddling them with records, destabilising families, and concentrating all of this geographically (see 8.6.2 on concentration effects). The prison became, for some communities, an institution of stratification — one of the mechanisms by which disadvantage is transmitted, sitting alongside the school and the housing market of Part 8.

Three things this analysis must not do.

It must not deny differential offending. Some crime is genuinely more common in conditions of concentrated disadvantage, for the structural reasons of 10.1.2 and Part 8 — and pretending otherwise is both false and patronising, treating the communities most harmed by crime as though their victimisation did not exist. The people most victimised by crime are disproportionately the same poor and minority communities most affected by punishment — which is why "less policing" and "less crime" are both demands made from within those communities, and the tension between them is real, not a confusion to be dissolved.

It must not treat the whole disparity as discrimination. The evidence is that both differential offending and differential treatment operate; assigning the whole gap to either is contradicted by the data (see 7.6.2). The honest estimate is partitioned and uncertain, not total in either direction.

And it must not lose the deviance of the powerful (see 10.1.1). The crimes of the advantaged — corporate fraud, financial crime, environmental crime, workplace deaths — cause more measurable harm than street crime and are punished far less reliably, less severely, and less often with prison. A sociology of punishment that studies only the punishment of the poor has reproduced the system's own selectivity. The most important fact about who gets punished is not only that the poor are over-punished but that the powerful are under-punished for larger harms — and the two are the same fact seen from opposite ends.

Because the punishment of crime is where the inequalities of Part 8 are enforced with the full authority of the state, and where the labelling of 10.1.3 becomes a life sentence in the literal sense.

Four questions to carry.

Is the disparity in offending or in treatment? For any punishment gap, ask how much is differential behaviour and how much differential processing — and demand a design that breaks the circle (self-report, hit rates, quasi-experiments), because raw comparisons and naive controls both mislead (see 8.4.2).

Does the control launder earlier disadvantage? Controlling for prior record is standard and removes part of the cumulative differential treatment from the estimate. A "justified" disparity may be built on an unjustified one upstream.

Does the account include the deviance of the powerful? A picture built only from who is in prison is a picture of who the system selects — 7.3.2's survivorship problem, with the powerful as the survivors who were never processed.

And is the institution tracking crime or policy? Mass incarceration rose while crime fell, which means it was a choice, not a response — and choices are reversible (see 9.7.1).

One closing observation, which draws the Topic together. Start from the two drug users. Identical behaviour, and one life is untouched while the other is organised around a record. The difference was not in what they did but in who they were and where they were — in the neighbourhood that was policed, the market that was available, the demeanour that was read, and the record that was written.

That is the sociological imagination applied to punishment (see 1.1.1): the outcome that looks like justice — this person is a criminal, that person is not — is substantially the product of how the response was distributed, and the response was distributed by the inequalities of Part 8. The prison is full of people who did things; it is also, and inseparably, full of people who were available to be caught doing them — and telling how much of it is which, honestly and without letting politics decide the answer, is the whole discipline of this lesson.

Two people using the same drug at the same rate: the advantaged one is never labelled, the disadvantaged one's identical use becomes the organising fact of his life — and self-report surveys show drug use is roughly equally distributed while arrest and imprisonment for it are wildly concentrated, proving that this punishment disparity is not produced by an offending disparity.

The punished population is everywhere disproportionately poor, young, male and — in racialised societies — from subordinated groups; the disparity is not in dispute, its cause is. Two families of explanation both operate: differential offending (real behavioural differences flowing from the structural conditions of Part 8) and differential treatment (unequal processing at the gates).

Disentangling them is hard because offending is confounded with everything and the measure of offending is itself system output — controlling for it is circular (8.4.2). The best evidence breaks the circle stage by stage: hit-rate outcome tests find lower contraband rates among more-stopped groups (evidence of a lower stopping threshold); charging and plea studies find disparities surviving controls; and sentencing disparities shrink but do not vanish after controlling for offence and record — with the crucial subtlety that prior record is itself laundered earlier differential treatment. The stage disparities are individually modest and cumulatively large because they compound.

Mass incarceration — the roughly fivefold US increase from the 1970s — was driven by policy, not crime: imprisonment rose while crime fell, so the series diverge. Harsher sentencing, the war on drugs and longer sentences were choices, reversible ones. Its consequences concentrate — becoming a modal life event for some groups, a collective condition reshaping families and neighbourhoods — so the prison became a producer of disadvantage and an institution of stratification, rejoining the sociology of punishment to the sociology of inequality.

Three cautions: do not deny differential offending (the most victimised communities are the most punished, and both demands come from within them); do not treat the whole disparity as discrimination (both operate, the estimate is partitioned and uncertain); and do not lose the deviance of the powerful (larger harms, punished less reliably — the same fact as over-punishment of the poor, seen from the other end).

Differential offending — the part of a punishment disparity produced by real differences in behaviour, themselves structurally caused.

Differential treatment — the part produced by unequal processing of the same behaviour.

The circularity problem — using system-generated offending measures to control for offending, which removes discrimination from the estimate.

Outcome (hit-rate) test — inferring a lower stopping threshold from a lower contraband-find rate among the more-stopped.

Cumulative disadvantage — the compounding of modest disparities across the five gates into a large total.

Laundering of prior record — the conversion of earlier differential treatment into a later "justified" disparity via the control for record.

Mass incarceration — the policy-driven expansion of imprisonment, diverging from crime.

Modal life event — an outcome so common in a group that it becomes typical rather than exceptional.

Prison as an institution of stratification — incarceration as a producer and transmitter of disadvantage.

Deviance of the powerful — larger harms by advantaged actors, punished less reliably; the mirror of over-punishment of the poor.

One — run the two users. Take an offence measurable by self-report and compare its distribution across groups with the distribution of punishment for it. The gap between the two is differential treatment.

Two — break the circle. For a claimed punishment disparity, ask whether the offending measure is independent of the system or produced by it. If produced by it, controlling for it hides the answer.

Three — trace the compounding. Take a small disparity at each of five stages and multiply them. The cumulative figure is much larger than any stage.

Four — check crime against prison. Plot your country's crime rate and its imprisonment rate over forty years. Where they diverge, imprisonment was tracking policy, not crime.

Five — find the under-punished. Identify one large harm caused by powerful actors that is rarely met with prison. Ask what the punishment statistics would look like if it were.