Is Mass Incarceration Mainly a Product of the War on Drugs?

At the end of 2023, state and federal authorities had jurisdiction over 1,254,200 people. States held about 1.1 million of them. The federal Bureau of Prisons held about 156,600, which is 12% of the total. Local jails held another 664,200 people, and the combined figure for adults in the custody of prisons or jails was 1,833,700.

Almost every intuition the public has about American prisons comes from federal data, and federal prisons are one prisoner in eight.

The two columns describe different countries. Federal prison is a drug, weapons and immigration system with almost no violence in it. State prison, which holds seven times as many people, is majority violent and one-eighth drug. Of that drug eighth, 37,790 people were serving time for possession, which is 3.6% of all sentenced state prisoners. The rest were trafficking or other commercial drug convictions. Women are about twice as likely as men to be in state prison for drugs, 26% against 12%.

So the arithmetic of the release-everyone question is not hard. Roughly 131,000 people in state prison plus roughly 65,000 in federal prison is about 196,000 people, a cut of about 16% to the prison population, leaving about 1.06 million. Add the jail share and the combined cut is around 15%, leaving well over 1.5 million people locked up.

That is the load-bearing number, and it has been stable across every version of this exercise anyone has run. FiveThirtyEight got a 14% cut on 2015 data. The Urban Institute, modelling 15 states, found that halving drug admissions would shrink those systems by about 7% over five years. The Sentencing Project, which is an advocacy organisation and says so, has made the same point from the other direction: drug counts fell 31% from their 2007 peak through 2018 and the prison population did not halve.

One more figure worth holding. In 1970 about 30% of state prisoners had been convicted of violence. By 2022 it was 63%. The state system did not become a drug system over the last half century. It became a violence system.

Everything above counts people on a single day. That measure systematically favours long sentences, and drug sentences are short.

A person serving nine months occupies a prison bed for a quarter of the year. A person serving seventeen and a half years occupies it for seventeen and a half. Count on any given day and the second one is there and the first almost certainly is not.

Count admissions instead and the picture moves. Jonathan Rothwell, working through the Bureau of Justice Statistics' National Corrections Reporting Program data for 1993 to 2011, found drug crimes accounted for about 31% of prison admissions against about 25% for violent crimes. In every year from 1993 to 2009, more people were sent to prison for drug crimes than for violent ones. That calculation is an analyst's work on BJS microdata rather than a BJS headline figure, which is worth knowing before quoting it.

Below the prison admissions is the jail churn, and it is enormous. Local jails processed 7.6 million admissions in the year to midyear 2023, down from 11.5 million in 2014 and still a flow that dwarfs everything else in the system. Seventy per cent of the people in jail on a given day have not been convicted of anything.

Below that again is arrest. More than 1.5 million drug arrests in 2019, about one in ten of all arrests in the country, 87% of them for possession. Rothwell counted roughly 30 million drug arrests between 1993 and 2011, 24 million of them for possession, against about 3 million prison admissions for drugs over the same period.

The great majority of people the drug war touches are arrested, booked, held for days or weeks, and released with a record. They never appear in a prison count. Someone arrested four times for possession over a decade has experienced incarceration four times and is not a drug prisoner in any table on this page.

The New Jim Crow made a version of this claim close to received wisdom, and it is usually summarised badly by both its supporters and its critics.

Michelle Alexander's thesis is about a system of racialized social control rather than a pie chart of prison beds. Her argument is that after the civil rights era, facially race-neutral criminal law reconstituted the work of Jim Crow, and that drug enforcement was the entry point. The durable part is not the sentence. It is the felony label, which legally licenses discrimination in employment, housing, education, public benefits, voting and jury service for the rest of a person's life.

On that ground the evidence supports her. Around 19 million Americans had a felony record as of 2010, about 8% of the adult population and about a third of Black adult men. Roughly 4 million Americans were disenfranchised by a conviction as of 2024, down from about 5.2 million in 2020 as states restored voting rights. Federal prisons really did become a drug system. Enforcement really was concentrated in poor urban neighbourhoods. The crack and powder disparity really was extreme.

One specific empirical claim does not hold. Alexander writes that drug convictions accounted for more than half of the increase in the state prison system between 1985 and 2000. Sentenced state prisoners rose from about 452,000 to about 1,206,000 in that window, an increase of roughly 754,000. Drug prisoners rose from about 39,000 to about 251,000, an increase of roughly 212,000. That is about 28% of the growth. The parallel claim about the federal system, that drug convictions drove about two thirds of its growth, is roughly right.

James Forman Jr. made this argument in the NYU Law Review in 2012. A theory that concentrates almost entirely on drug crime, he wrote, cannot explain a system in which violent offenders are about half the prisoners and drug offenders about a quarter. Forman is not an opponent of decarceration. He went on to write a book about how Black political actors helped build punitive policy in Washington DC, which is a further complication for any single-cause account.

Alexander's available defence is that she was describing a mechanism and a political permission structure, not a spreadsheet. That is a real argument and it is much harder to falsify, which is a reason to state it as the different claim it is rather than as a fact about who is in a cell.

This is where the drug-war case is strongest and where its most important limitation sits, about four paragraphs apart.

Survey data have shown rough parity in drug use by race for decades. In the 2023 National Survey on Drug Use and Health, past-year illicit drug use was 26.1% among white Americans, 27.7% among Black Americans, 21.6% among Hispanic Americans, 36.7% among American Indian and Alaska Native Americans and 12.4% among Asian Americans. For marijuana specifically it was 23.1% and 24.5%.

Arrests do not follow that pattern. The ACLU, using 2018 FBI data, found Black Americans 3.64 times more likely than white Americans to be arrested for marijuana possession, with disparities in every state and in more than 96% of the counties it examined. In 2022, Black Americans were about 14% of the population and 27.3% of arrests for drug abuse violations where race was recorded. Going back further, a Bureau of Justice Statistics comparison for 1991 to 1993 found Black people were 40% of drug arrestees against 13% of self-reported drug users, and 13 percentage points of the gap remained after adjusting for the race-neutral factors it could measure.

Charging adds more. Rehavi and Starr, studying federal cases from 2007 to 2009, found a Black-white sentence length gap of about 10% surviving controls for arrest offence and criminal history, more than half of it traceable to the initial charging decision, with prosecutors filing charges carrying mandatory minimums against Black men about 65% more often. The Anti-Drug Abuse Act of 1986 set the crack to powder quantity ratio at 100 to 1, and around 80% of crack defendants were Black against around 30% of powder defendants. The Fair Sentencing Act cut the ratio to 18 to 1 in 2010 and the First Step Act made that retroactive in 2018, producing 3,705 sentence reductions. The 18 to 1 ratio is still there.

Now the part that cuts the other way, from the same BJS table that produces the 62% figure.

Black and Hispanic state prisoners are less likely than white state prisoners to be there on a drug conviction, and considerably more likely to be there for violence. Black Americans are overrepresented in every offence category relative to their population share, rather than concentrated in drugs.

The consequence is arithmetic. John Pfaff calculates that releasing everyone in prison whose top charge was a drug offence would move the Black share of prisoners from 38% to 37%. Heather Mac Donald, running the same exercise on BJS offence-by-race tables, gets 37.4% to 37.2%. The Council on Criminal Justice found the Black-white disparity in drug imprisonment fell 75% between 2000 and 2019, accounting for about half of the entire fall in the imprisonment disparity over those years, and that the largest remaining gap is in violent offences, where Black adults were imprisoned at more than six times the white rate.

Both things are true at once. Drug enforcement is applied unequally by race at the point of arrest and at the point of charging, by margins that are large and well documented. And ending drug imprisonment tomorrow would leave the racial composition of American prisons almost exactly as it is.

The best-argued alternative comes from John Pfaff, and the first thing to say about it is that Pfaff wants fewer people in prison. So do Forman, the National Academies committee, the Sentencing Project and the Prison Policy Initiative. The fight is about the route, and reading it as reformers against defenders of prison gets it backwards.

Pfaff's argument in Locked In covers roughly 1994 to 2008. Crime fell. Arrests fell. The probability that a filed felony case ended in a prison admission stayed roughly flat. The thing that rose was felony filings in state courts, by about 40%, matching a 40% rise in prison admissions. The number of line prosecutors grew about 50% between 1990 and 2007 while crime was falling. His reading is that prosecutors became more willing to file felony charges against the arrestees in front of them, and that county prosecutors face local political incentives while the cost of a prison bed falls on the state. On his decomposition of 1990 to 2009 state confinement growth, violence contributed about 60% and drugs about 14%.

The measurement underneath that is contested, and the dispute is technical rather than ideological.

What is not in dispute is that drug headcounts cannot explain the growth of the 1990s and 2000s. The National Academies' 2014 synthesis concluded the fivefold rise from 1973 was preponderantly the result of increases in both the likelihood of imprisonment and the length of sentences, with sentence length the primary cause since 1990. Mandatory minimums, truth-in-sentencing, three-strikes laws, life without parole and the abolition of discretionary parole did that work, and they were applied across the board. Prison commitments per 100 adult arrests rose between 1980 and 2010 for drugs, from 2 to 9. They also rose for murder, from 41 to 92, for sexual assault from 8 to 30, and for aggravated assault from 4 to 14.

Drugs were the opening wedge for that machinery. The machinery was then used on everything.

Portugal decriminalised acquisition, possession and use of all drugs up to a ten-day personal supply in July 2001, and referred those cases to administrative commissions instead of courts. Trafficking remained a crime. It is the closest thing to a natural experiment anyone has, and it is cited constantly by people who have not looked at the prison series.

It dipped, then came back to roughly where it started. The reason is the interesting part. Before 2001, fewer than 1% of people incarcerated in Portugal for a drug offence had been convicted of use. Hannah Laqueur's account puts the number of people in prison for drug use at around 25. The law largely wrote down what Portuguese courts were already doing.

What did change was composition. The drug-offence share of the prison population fell from a peak near 44% in 1999 to the mid-teens by 2019, and convictions and prison sentences for trafficking fell even though trafficking arrests did not. Public health outcomes improved substantially, and HIV among people injecting drugs fell sharply. Those are real results and they are not prison-population results.

Decriminalising use did not empty Portuguese prisons because use was not filling them. The same thing should be expected of American cannabis legalisation, and roughly the same thing happened: marijuana possession was the controlling offence for about 1.1% of American inmates, so removing that category can eliminate hundreds of thousands of arrests a year while barely moving a prison count.

On the international comparison, the United States incarcerates at roughly 530 to 540 per 100,000 counting prisons and jails together. Germany is around 70, the Netherlands around 60 to 70, France around 105, Spain around 121, England and Wales around 135, Portugal around 118, Norway around 54 and Japan around 38. Remove every drug prisoner in America and the rate stays several times Germany's. Nobody has published a decomposition splitting the gap between drug policy and the length of American sentences for violence, and the direction of the evidence is that the second is doing most of the work.

If drug enforcement was the engine, the decline since the peak is a direct test of it.

The sentenced state and federal prison population peaked at about 1.6 million in 2009 and stood at 1.25 million at the end of 2023, about 22% below peak and 2% above 2022. Across 39 states with comparable data, Pew found drug prison admissions fell 34% and the drug prison population fell 33% between 2009 and 2019, against a 24% fall in total admissions and a 10% fall in the total prison population. The shrinking drug population accounted for 61% of the overall decline and 38% of the decline in admissions.

So drug reform delivered more than its share of the decline. That is the strongest evidence on this page that drug policy is a high-leverage lever, and it is descriptive rather than causal.

Two details complicate it. Over the same decade, arrests for drug possession fell by less than half of one per cent. The front end of the system did not change at all; what changed was what happened to people after they were arrested. And the Council on Criminal Justice found that the fall in drug imprisonment disparity came through changes in practice rather than changes in sentencing law, with legislation largely recording decisions police and prosecutors had already made.

The national figure also conceals most of what happened. Drug convictions were 24% of prison admissions across 43 reporting states in 2019, ranging from 10% or less in Alaska, California, Colorado and Vermont to 40% or more in Arkansas, Kentucky and South Dakota. Of the 39 states with comparable data across the decade, 11 increased their drug prison populations. State imprisonment rates run from about 660 per 100,000 in Mississippi and about 600 in Louisiana down to about 95 in Massachusetts, a sevenfold spread inside one country. The 2020 collapse was court closures and emergency releases rather than reform, and the rebound through 2021 to 2023 is the evidence for that.

The strongest objection to every table on this page is that the labels are wrong. A robbery committed to fund a habit is a violent prisoner in BJS data, and a drug conviction may be a plea from something else. The last comprehensive surveys of American inmates, conducted between 2002 and 2004, found 75.9% were substance-involved on a broad definition, and 77.5% of those whose controlling offence was violent. Marijuana possession was the controlling offence for 1.1% of them. Nobody has worked out how many of those robberies happen in a country with different drug laws, and no dataset anyone has collected can answer it.