County-level decarceration atlas: mechanisms, prevalence, and dynamics of decarceration across 2,870 U.S. counties, 1999-2019

This paper presents a national "atlas" characterizing the mechanisms, prevalence, and dynamics of decarceration across 2,870 U.S. counties from 1999 to 2019, identifying four distinct operational types and revealing significant sub-state heterogeneity to serve as a framework for studying decarceration's health effects.

The Gist

Imagine the United States as a giant, bustling city with thousands of neighborhoods (counties). For decades, this city has been building more and more "holding cells" (jails and prisons) and keeping people inside them for longer and longer. This is what we call mass incarceration.

Recently, people have started asking: "What happens if we start letting people out?" This process is called decarceration. But here's the problem: we don't really know if letting people out is good or bad for the health of the community, because every neighborhood is doing it differently. Some are letting people out because they changed the laws; others are doing it because they ran out of money; some are just releasing people earlier, while others are stopping new people from entering.

This paper is like a giant, detailed map (an "Atlas") that the authors drew to show exactly how 2,870 different neighborhoods across the U.S. started letting people out between 1999 and 2019.

Here is the breakdown of what they found, using some simple analogies:

1. The Four Ways People "Get Out"

The authors realized that "letting people out" isn't just one thing. They sorted it into four different "doors" or mechanisms:

• The "Front Door" (Reduced Pretrial Detention): This is like stopping people from even entering the building. Instead of locking someone up while they wait for their court date, the neighborhood decides to let them stay home.
• The "Shorter Stay" (Reduced Jail Time): People are still entering the building, but they are leaving the jail much faster. Imagine a hotel where guests used to stay for a month, but now they only stay for a week.
• The "No New Guests" (Reduced Prison Admissions): This is about the big, long-term prisons. The neighborhood decides to stop sending new people there. The "No Vacancy" sign is up.
• The "Early Release" (Reduced Prison Time): People are already inside the long-term prison, but the neighborhood is letting them out earlier than planned.

2. What the Map Showed

The authors looked at the data and found some surprising things:

• It's Happening Everywhere: About two-thirds of all neighborhoods (including many rural ones) tried at least one of these "getting out" strategies. It wasn't just big cities; small towns were doing it too.
• It's Often a "Reaction," Not a "Plan": Think of it like a thermostat. For years, the "heat" (incarceration rates) kept rising. Eventually, the neighborhoods got so hot (overcrowded, expensive, or legally pressured) that they finally turned the AC on. Most of the time, decarceration wasn't a proactive choice to be kind; it was a reaction to the system getting too full.
• The Drops Were Modest: When neighborhoods started letting people out, the numbers didn't crash to zero. They dropped by about 20% to 40% over ten years. It was a slow leak, not a flood.
• Every Neighborhood is Different: Just like no two families eat the same dinner, no two counties decarcerated the same way.
  • Big cities often stopped sending people to prison.
  • Rural areas had different patterns.
  • Sometimes, a whole state passed a law to let people out, but only some counties in that state actually did it. This shows that local judges and police chiefs have a lot of power, even when the state says "stop."

3. Why This Map Matters

Imagine you are a doctor trying to figure out if a new medicine works. If you give the medicine to 1,000 people, but 500 take it with food, 300 take it on an empty stomach, and 200 take a different brand entirely, you can't tell if the medicine works. You need to know how they took it.

This paper does the same thing for the criminal justice system.

• Before: We knew incarceration was bad for health.
• Now: We know that "letting people out" isn't just one thing.
  • If you stop admitting people to prison, fewer people will be released later (which might lower overdose rates, as release is a high-risk time).
  • If you just let people out earlier without stopping admissions, you might suddenly have a huge wave of people returning to the community who need support.

The Bottom Line

This "Atlas" is a tool for future researchers. It says: "Don't just ask, 'Did we let people out?' Ask, 'How did we let them out, and what happened next?'"

By understanding the specific "mechanism" (the door they used), we can finally figure out which ways of reducing incarceration actually make communities healthier and safer, and which ones might accidentally cause new problems. It's the first step toward fixing a broken system in a way that actually helps people.

Technical Summary

1. Problem Statement

The United States possesses the highest incarceration rate globally, with documented adverse effects on population health and health equity. While "decarceration" (reducing incarceration rates) is increasingly viewed as a public health strategy, the evidence regarding its health effects remains limited and inconclusive.

• Knowledge Gaps: Existing literature often relies on periods of rising incarceration or focuses on specific state-level policies that are difficult to scale.
• Mechanism Ambiguity: It is unclear how decarceration occurs (e.g., reduced admissions vs. reduced time served) and whether different operational mechanisms yield different health outcomes.
• Data Limitations: There is a lack of national, county-level data characterizing the specific mechanisms, timing, and magnitude of decarceration episodes across heterogeneous local contexts.

2. Methodology

The study utilized a longitudinal, ecological design to map decarceration dynamics across the U.S.

Data Sources:

• Incarceration Data: County-level jail and prison data from the Vera Institute of Justice (1999–2019).
• Population Data: U.S. Census Bureau data for the resident population aged 15–64 (used as the denominator to approximate the population at risk).
• Geographic Scope: 2,870 U.S. counties. States with unified prison/jail systems were excluded due to data unavailability.
• Variables: Four key measures: pretrial jail population, total jail population, prison admissions, and prison population.

Analytical Approach:

• Typology Definition: The authors defined four operational types of decarceration based on trends in the four measures:
  1. Reduced Pretrial Detention: Decrease in pretrial jail population.
  2. Reduced Jail Time: Total jail population decreases more steeply than pretrial population (indicating shorter sentences for those detained).
  3. Reduced Prison Admissions: Decrease in the flow of individuals entering prison.
  4. Reduced Prison Time: Decrease in the prison population relative to admissions (indicating shorter time served).
• Trend Detection:
  • Applied piecewise negative binomial regression for each county and measure.
  • Models included zero, one, or two breakpoints; the best fit was selected using the Akaike Information Criterion (AIC).
  • Classification Criteria: A segment was classified as "decarceration" if it had a negative slope with an adjusted $p$-value $\le 0.05$ (Benjamini-Hochberg correction), a cumulative decline of $>10$ individuals, and did not occur at the extreme edges of the observation window (to avoid edge effects).
• Validation: Construct validity was assessed by comparing results against eight well-documented case studies (state reforms, local ordinances, fiscal constraints) to ensure the algorithm identified expected mechanisms and timing.

3. Key Contributions

• First National Atlas: This study provides the first systematic, county-level characterization of decarceration mechanisms across the U.S. over a 20-year period.
• Operational Framework: It moves beyond binary "incarcerated vs. not incarcerated" metrics to distinguish between "front-end" (admissions) and "back-end" (time served) reductions in both jail and prison systems.
• Scalable Resource: The resulting "atlas" serves as a hypothesis-generating tool for future comparative health studies, allowing researchers to link specific decarceration mechanisms to health outcomes (e.g., overdose mortality, infectious disease transmission).
• Revealing Heterogeneity: The study highlights that state-level reforms often result in highly variable implementation at the county level, driven by local discretionary actors (judges, prosecutors).

4. Key Results

Prevalence and Frequency:

• Coverage: Approximately 65% of counties with both jail and prison data experienced at least one type of decarceration between 1999 and 2019. This includes most rural counties.
• Most Common Type: Reduced prison admissions (33% of counties), followed by reduced prison time (29%), reduced jail time (25%), and reduced pretrial detention (25%).
• Urbanicity: Decarceration was more frequent in large central metropolitan counties (90% experienced reduced prison admissions) compared to rural counties (23%).
• Timing: Decarceration expanded significantly in the 2000s, peaking around 2008–2009, and then leveled off. Large metropolitan areas continued to see expansion into the 2010s, while other regions plateaued or declined.

Magnitude and Dynamics:

• Decline Size: Median reductions ten years after onset ranged from 19% to 38%:
  • Reduced pretrial detention: 37%
  • Reduced prison admissions: 38%
  • Reduced jail time: 24%
  • Reduced prison time: 19%
• Trajectory: Declines were steepest in the first 5–6 years and then plateaued. Most declines were not reversed.
• Reactive Nature: Archetypal trajectories revealed that decarceration episodes typically followed periods of increasing incarceration rates, suggesting they were often reactive adjustments to overcrowding or budget strain rather than proactive, sustained policy shifts.

Validation:

• The algorithm successfully aligned with known events, such as California's 2011 Public Safety Realignment (surge in reduced prison admissions) and North Carolina's Justice Reinvestment Act (shift from reduced admissions to reduced time served).
• It also detected significant intra-state heterogeneity, showing that many counties within reforming states did not experience decarceration, continuing to see rising rates.

5. Significance and Implications

• Health Research Foundation: The atlas establishes a descriptive baseline necessary to study the heterogeneous health effects of decarceration. It suggests that the mechanism of decarceration matters; for example, reduced prison admissions (fewer people entering) may have different overdose mortality implications than reduced prison time (more frequent releases of the same number of people).
• Policy Nuance: The findings challenge the notion of a uniform "decarceration" trend. Instead, they reveal a complex landscape of local adjustments, fiscal constraints, and discretionary practices.
• Future Directions: The study calls for future research to:
  • Compare health outcomes across different decarceration types.
  • Investigate effect heterogeneity based on local context (e.g., availability of community treatment).
  • Explore the role of local actors in mediating state-level reforms.
  • Expand analysis to include community supervision, which was not captured in this dataset but is a growing area of carceral control.

Conclusion:
While decarceration is occurring across the majority of U.S. counties, it is often modest in magnitude and reactive in nature. This atlas provides the necessary granularity to move beyond broad generalizations, enabling researchers and policymakers to understand how specific operational changes in the criminal legal system impact population health and equity.