Skip to content
Read • Question • ConnectAnthony’s notes & researchReturn with a new question
Anthony’s NotesA personal field guide
From the notebook

The Unheralded End of Mass Incarceration

⬅️ Hub: Crime · Related: Felonies in Abundance · Repeat Offenders Crime Concentration · The Hidden Civic Tax of Crime

Bottom line

Prosecutors can make charges, felony labels, convictions, and prison admissions fall by declining cases, reducing charges, expanding diversion, or recommending shorter sentences. The best recent estimate finds reform prosecutors reduced misdemeanor charges by about 19%, felony charges by 22%, and convictions by roughly 23–25% relative to comparison jurisdictions. That is a large effect on court-output statistics.

It does not follow that the broad fall in violent and property crime is a prosecutorial accounting trick. FBI/NIBRS offenses are recorded by police before a DA acts, and the National Crime Victimization Survey counts reported and unreported victimization independently of police and prosecutors. A 2026 peer-reviewed decomposition finds the 2000–2019 declines in violent- and property-crime imprisonment were driven almost entirely by fewer recorded crimes, while the decline in drug imprisonment was largely an enforcement and case-processing change. The defensible conclusion is therefore: the artifact concern is important for charges, convictions, drug enforcement, and incarceration, but it does not explain most of the national violent/property crime decline.

Source and status

  • Article analyzed: Keith Humphreys, Shawn D. Bushway, and Jason Robey, “The Unheralded End of Mass Incarceration,” Washington Monthly, September 30, 2026.
  • What it is: interpretive magazine commentary by the authors of the research it discusses—not an official statistical release.
  • Research anchor: Bushway, Robey, and Humphreys, “Running Out of Inmates? Cohort Nonreplacement and the Approaching End of Mass Incarceration,” Annual Review of Criminology (review in advance, September 25, 2026).
  • Evidence cutoff: September 30, 2026. Mid-2026 crime estimates are preliminary; the latest nationally comparable court-processing series remain much older and fragmented.

What the article argues

The article says that mass incarceration is unwinding because younger birth cohorts are less likely to commit crime, be arrested, enter prison, or replace older incarcerated cohorts. It treats this as a demographic and behavioral change more than a simple result of sentencing reform. Its strongest claims are that prison admissions will keep falling, imprisonment among younger adults has already fallen sharply, the prison population is well below its 2009 peak, and racial incarceration gaps have narrowed.

That overall thesis is plausible and substantially supported, but several headline formulations need tighter denominators and dates.

Article statistics audit

Article claimBest available checkAssessment
Prison admission rates will fall another 50% over the next decade.The linked Annual Review abstract projects an additional 50% within two decades.Direction supported; the article compresses the stated horizon from two decades to one.
The U.S. prison population is down a quarter from its 2009 peak.BJS reports 1,615,487 prisoners at the 2009 peak and 1,254,200 at year-end 2023: a 22.4% decline.Close in round-number prose, but the latest fully published BJS count supports about 22%, not exactly 25%.
The Black incarceration rate is less than half its early-century level.For Black men, the imprisonment rate fell from 3,547 per 100,000 in 2000 to 1,862 in 2023, a 47.5% decline; for Black women, 205 to 68, a 66.8% decline.Broad direction is right; the exact fraction depends on sex, population definition, and endpoint.
Under-30 prison admission rates have been cut roughly in half.The authors’ cohort analysis supports very large declines among younger cohorts; youth arrests fell 84% from 1996 to 2020, and delinquency cases involving detention fell 66% from 2005 to 2022.Strong directional support, but these are different measures and should not be treated as one interchangeable series.
Mid-2026 homicide is at a century-scale low.The Council on Criminal Justice found homicides down 18% in the first half of 2026 versus 2025 and 51% from the 2022 peak across 36 cities; its national record-low statement is a projection pending later FBI data.Encouraging but preliminary and not yet a final national count.

Visual audit charts

Each interactive chart carries its source citation in the footer. The first and fourth cite the linked Bushway, Robey, and Humphreys study because the article draws those claims from that research.

Projection horizon: article versus linked study

Open the interactive projection-horizon chart

Prison population: 2009 peak versus 2023

Open the interactive prison-population chart

Black imprisonment rates: men and women

Open the interactive Black-imprisonment-rate chart

Youth measures: cohort-adjacent evidence

Open the interactive youth-measures chart

The statistical pipeline

flowchart LR
    A[Victimization and offending] --> B[Police-recorded incidents]
    B --> C[Arrests]
    C --> D[Matters reviewed by prosecutors]
    D --> E[Charges filed]
    E --> F[Convictions or diversion/dismissal]
    F --> G[Prison admissions]
    G --> H[Prison population]

Every stage answers a different question:

MeasureWhat it countsCan a DA directly lower it?Main caveat
NCVS victimizationSurveyed violent and property victimization, whether or not reportedNoExcludes homicide, business crime, people without households, and institutional populations.
FBI/NIBRS offensesIncidents known to and recorded by law enforcementNoReporting, police recording, agency participation, and definitions can change.
ArrestsArrest events reported by policeUsually no; indirectly possibleOne person can be arrested more than once; enforcement intensity changes counts.
Prosecutor matters/chargesCases reviewed and formally filedYesOffices count matters, defendants, incidents, and charges differently.
ConvictionsGuilty pleas and trial convictionsYesA felony arrest can become a misdemeanor conviction; diversion and dismissals are not acquittals.
Prison admissions/populationEntries into prison and people confinedYes, with judges and legislaturesPopulation is a stock shaped by both admissions and time served.

This is why a drop in convictions cannot by itself prove that crime fell, but it also cannot make NCVS victimization or already-recorded police incidents disappear. The possible indirect pathways are real but harder to measure: police may make fewer arrests when they expect declination; victims may report less if they expect no action; and drug “crime” is particularly enforcement-generated because there is usually no complainant.

Arrests, charges, and convictions over time

There is no single, clean national annual series that follows the same incidents from arrest through charge and conviction. Police data are national and frequent; prosecutor and state-court data are decentralized, use inconsistent counting units, and have large gaps. The best evidence is therefore a set of bounded snapshots and series rather than one seamless line.

Stage and periodChange or dispositionWhat can safely be concluded
FBI arrests, 2010–2019Total arrests −21.0%; violent-crime arrests −7.8%; property-crime arrests −29.6%; juvenile arrests −55.5%; adult arrests −16.0%.Arrests clearly fell before the pandemic. This FBI trend table includes only agencies reporting 12 months in both years and counts arrest events, not unique people.
State felony convictions, 1990–2006829,340 → 1,132,290, up 36.5%; the rate rose 447 → 503 per 100,000 adults.Convictions rose during the late mass-incarceration expansion. This BJS series ends in 2006 and does not establish what happened afterward.
Large urban felony cases, 1990–2009Among cases already filed and adjudicated, any-conviction rates were generally 64–68%; felony-conviction rates 50–58%. In 2009: 66% convicted, 25% dismissed, 1% acquitted, and 9% diverted/deferred.The post-filing conversion rate was comparatively stable through 2009, even though the volume and composition of arrests could change. About 18% of 2009 convictions arising from felony charges were misdemeanor convictions.
State prosecutors, 2020 snapshotOf 2,749,600 felony matters reviewed, 78% were filed and 22% declined. Of 1,872,880 concluded felony cases: 66.7% ended in a plea, 17.1% dismissal, 4.5% trial, and 11.8% other outcomes such as diversion or problem-solving courts.Declination and dismissal are quantitatively important, but 2020 was a pandemic-disrupted year and this survey is not directly comparable with earlier BJS reports.
Reform-prosecutor jurisdictions, modern event-study estimateMisdemeanor charges −19%, felony charges −22%, misdemeanor convictions −23%, felony convictions −25% relative to comparison jurisdictions.DA policy can produce large changes in downstream case statistics even when reported incidents show no comparably large change. This is a March 2026 working paper, not yet a final peer-reviewed estimate.

A historical arrest-to-conviction funnel

A BJS study of felony arrests in 30 urban jurisdictions in 1988 illustrates ordinary attrition long before the current progressive-prosecutor movement. Out of 100 adult felony arrests, about 18 were rejected before filing, 21 were dismissed after filing, 6 were diverted or otherwise referred, 1 ended in acquittal, and 54 ended in conviction. Of those convictions, 52 resulted from guilty pleas and 2 from trial; 32 of the original 100 arrests led to incarceration.

An older 1981 study reported 48 convictions per 100 felony arrests, with 23 rejected, 22 dismissed, 6 diverted/referred, and 1 acquitted. The apparent 48-to-54 increase should be treated as suggestive, not a national time trend: the jurisdictions and data systems differed. The key point is that large gaps between arrest and conviction are not new, and a dismissal is not automatically proof that a prosecutor ignored a provable crime. Evidence, legal sufficiency, duplicate charges, witness availability, diversion, and plea bargaining all affect the funnel.

How much do reform DAs change the numbers?

The strongest current direct estimate is the March 2026 working paper “Prosecutorial Reform and Local Crime Rates.” Using staggered differences-in-differences and Criminal Justice Administrative Records System case data, it estimates that after a reform prosecutor takes office:

  • misdemeanor charges fall by 622 per 100,000 residents per year, or 19%;
  • drug-misdemeanor charges fall by 243 per 100,000, or 42%;
  • the share of misdemeanor charges ending in conviction falls 9–11 percentage points, or roughly 17–24%, depending on offense type;
  • misdemeanor convictions fall by 333 per 100,000, or 23%;
  • felony charges fall by 335 per 100,000, or 22%;
  • the felony charge-to-conviction share falls 2.5 percentage points, or 5%;
  • felony convictions fall by 157 per 100,000, or 25%; drug-felony convictions fall 57%, and violent-felony convictions fall by about 24 per 100,000, or 25%;
  • the estimated fall in prison incarceration is about 5%, but is not statistically distinguishable from zero.

Those are substantial changes. They mean that a graph of felony charges, convictions, or prison admissions can fall partly because the system is processing conduct differently—not only because the conduct became rarer.

But the same paper finds no statistically significant broad change in police-reported incidents. Its confidence intervals generally rule out crime increases larger than about 10%, except drug incidents; it cannot rule out an approximately 7% property-crime increase. The authors explicitly warn that civilian reporting and police recording can respond to prosecutorial policy, so the estimate does not close every loophole in the artifact hypothesis.

The wider causal literature is mixed but does not support a general claim that reform prosecutors caused the national crime decline—or a general violent-crime increase:

  • A 2024 peer-reviewed county study estimated roughly 7% higher property and total reported crime after progressive-prosecutor inaugurations, driven by property crime, with no general violent-crime increase. Both progressive and traditional jurisdictions still experienced large absolute crime declines over the longer period.
  • A 2023 Quarterly Journal of Economics study exploited as-if random assignment in Suffolk County, Massachusetts. For marginal nonviolent-misdemeanor defendants, nonprosecution produced a 53% lower probability of a new criminal complaint and 60% fewer new complaints over two years. This result is not evidence about serious violent defendants, but it directly contradicts the assumption that every declined low-level case increases later crime.

Where prosecution policy did matter most: drug cases

Robey and Light’s 2026 peer-reviewed decomposition is the most direct answer to the article’s causal claim:

  • From 2000 to 2019, violent-crime prison admissions fell about 28%, property-crime admissions 40%, and drug admissions 46%.
  • Recorded violent crime fell about 32%; property crime reports fell 41%, from about 4,590 to 2,730 per 100,000 between 2005 and 2019.
  • For violent and property crime, arrest probability per recorded crime barely changed, while the chance of prison following arrest increased slightly. The imprisonment decline was therefore driven almost entirely by fewer recorded crimes, not lenient case processing.
  • Drug admissions were different: there was little evidence that drug use fell enough to explain the change. For Black Americans, the likelihood of drug arrest fell 46% and the likelihood of prison after a drug arrest fell 53%, producing a 75% decline in Black drug-prison admissions. White drug admissions fell 13%.
  • The Black–white disparity in drug prison admissions fell from more than 12:1 to less than 3:1. Marijuana admissions were only about 3% of drug admissions, so legalization alone cannot explain the shift.

This is the cleanest division of the evidence: real reductions in recorded violent and property crime explain most of those incarceration declines; enforcement and processing decisions explain much more of the drug decline.

Verdict on the “crime only looks lower because charges were dropped” hypothesis

What is true, false, and unresolved

True: DA policy can lower charges and convictions by about one-fifth to one-quarter, with still larger effects for drug cases. It can also move conduct from felony to misdemeanor, conviction to diversion, and prison to community sanctions. These downstream statistics must not be mislabeled as “crime.”

Not supported: The claim that the broad national decline in violent and property crime is mainly an illusion created by dropped or reduced charges. Police-recorded incidents and NCVS victimization are upstream measures, and the best 2000–2019 decomposition attributes the imprisonment decline in those categories to fewer recorded crimes.

Still possible in limited settings: Police effort, victim reporting, and offense recording may change when prosecution policy changes. This is most serious for drug and low-level public-order offenses, for which enforcement partly creates the count. Local property-crime effects remain uncertain and may be modest rather than zero.

Wrong inference to avoid: A falling conviction count does not prove falling crime; a falling prosecution rate does not prove rising crime; and a dismissal does not tell us whether the underlying allegation was true or provable.

What better data would answer the question

The ideal system would assign a stable incident identifier and follow the same event through victim report, police recording, arrest, prosecutor review, initial and final charges, diversion or dismissal reason, conviction, sentence, and later offending. It would publish this by offense, seriousness, prior record, race, age, and jurisdiction under consistent definitions. The United States does not currently provide that national longitudinal pipeline. BJS’s most recent large-urban felony-processing release still covers 2009; its newer collection is not yet a public replacement time series.

Until that exists, the most honest dashboard keeps at least four lines separate:

  1. NCVS victimization rates;
  2. FBI/NIBRS recorded-offense rates;
  3. arrest rates and arrest-to-offense ratios;
  4. prosecutor filing, conviction, and incarceration rates per arrest and per resident.

Sources