Recidivism alone cannot tell you whether reentry is working
Most reentry programs are judged in the end by one number: how many participants return to the justice system. That number matters, and nobody should stop reporting it. But it arrives years late, it counts only failure, and it says nothing about what happened to a person in the months that decided their path. A program officer who sees only recidivism learns, three years after the money is spent, whether the program worked. A program officer who also sees housing, work, documents, treatment and family connection learns it while there is still time to change something.
The outcomes worth measuring are the ones that describe a person building a life: a job, a place to live, an ID in their wallet, a doctor, benefits they qualify for, a family that is back in touch. Those are what the program is actually trying to produce. Measure them carefully, and they become the early read on whether the long-term number will move.
What recidivism actually counts
Before adding new outcomes, get the old one right. "Recidivism" is not one measure. Federal studies that follow released people over time report rearrest, reconviction and return to prison or jail as separate measures, and the three produce very different numbers for the same group. Rearrest is the broadest and comes first; reincarceration is narrower and can include returns for technical violations of supervision, such as a missed appointment, not just new crimes.
Then there is the window. A one-year rate and a five-year rate for the same cohort are different facts. A program that reports one-year rearrest and compares itself to a published three-year reincarceration figure is comparing two different things, and the comparison flatters or condemns it for reasons that have nothing to do with the program.
Write the definition down before the first participant enrolls: which event counts, from what date the clock starts, how long the window runs, and which data source records it. Then use that definition for every cohort, every year.
The outcomes that tell you whether reentry is working
Seven outcomes cover most of what matters in the first year after release. Each needs a precise definition, a fixed window, and a named data source, or it will be measured differently by every partner in the network.
| Outcome | Definition | Window | Data source |
|---|---|---|---|
| Employment | Earnings reported in a calendar quarter, or verified work of a set number of hours a week | Second and fourth quarter after release or enrollment | State unemployment insurance wage records, with pay stubs or employer verification for work those records miss |
| Stable housing | Living in a place the person can stay at least the next 90 days, not a shelter, the street, or a night-to-night arrangement | 30, 90 and 180 days | Case manager verification; homeless management information system records where available |
| Identification documents | Holds a state ID or driver's license, a Social Security card and a birth certificate | Within 30 days | Case manager verification of each document |
| Treatment engagement | For people with an identified need, attended a second treatment session after the first | Within 14 and 30 days of release | Provider records, shared with consent |
| Benefits enrollment | Enrolled in each benefit the person is eligible for, such as health coverage and food assistance | Within 30 and 60 days | Case manager verification; state agency data under an agreement |
| Family reconnection | Regular contact with a family member or support person the participant chooses, as the participant describes it | 90 and 180 days | Participant self-report on a standard question |
| Supervision compliance | For people under supervision, no violation filed in the window | 90, 180 and 365 days | Probation or parole records under a data agreement |
Two details in that table do most of the work. The employment window follows the calendar quarters that state wage records are reported in, which is how workforce programs commonly measure employment and why reentry programs that align with it can compare results. And treatment engagement is defined as the second session, not the first, because the first appointment is the one people are walked into, and the second is the one they choose.
Which outcomes predict lasting success
Not every outcome carries the same weight. In our experience, some early outcomes are strong signals that a person is on stable ground, and others are useful mainly because they make the strong ones possible. The bars below are our judgment, not a measured result, and they should be read as a starting point for a program's own evaluation.
Early Outcomes as Signals of Lasting Stability
Our judgment from practice, not measured data: relative strength of each outcome as an early signal that a person is on stable ground.
Housing and work sit at the top because nearly everything else depends on them. A person sleeping in a different place every night struggles to keep appointments, hold a job, or stay reachable by a case manager. Documents and benefits rank lower not because they matter less, but because they are means rather than ends: an ID is what makes the job application and the apartment lease possible. Supervision compliance ranks lowest as a signal because it reflects supervision practice as much as the person's own progress, a point the next sections return to.
Measurement windows that mean something
Every outcome needs a start date and a window, applied the same way to everyone. The start date is usually the release date or the enrollment date. Pick one and hold it. A program that enrolls people months before release and one that enrolls them on the day they walk out are measuring different stretches of a person's life, and both should say which one they mean.
Report by cohort, meaning everyone who started in the same quarter, and follow each cohort through the full window before reporting a final result. A rate computed on whoever happens to be enrolled right now mixes people in their first week with people in their tenth month and moves every time enrollment changes. Report interim results for cohorts still in their window as interim, and label them that way on the dashboard.
Count everyone who enrolled, not just everyone who finished. A program that reports outcomes only for completers measures who stayed, not what the program did. Report both, and put the enrolled number first.
Data sources and what each one misses
Each source sees part of a person's life and misses the rest. State unemployment insurance wage records are the most reliable source for formal employment, but they generally do not capture self-employment, informal work, or work in another state, and they arrive a quarter or more after the fact. Homeless management information systems record shelter and housing program stays, but not a person staying with family. Probation and parole records show supervision events, but each agency defines and records violations its own way.
Plan for the gaps rather than pretending they do not exist. Use administrative data where you can get it under an agreement, case manager verification for what administrative data misses, and a short participant survey at fixed points for the outcomes only a person can report, such as family connection. Record which source produced each data point, so an analyst can see when a change in the numbers is really a change in the source.
Consent and data sharing
Many of these outcomes depend on records held by other agencies and providers, and the participant's consent is what makes most of that sharing possible. Write the consent form in plain language, at a reading level that works for everyone the program serves. Say what will be shared, with whom, for what purpose, and for how long, and say how to withdraw consent. Explain it in person; a form signed in a rush on the day of release is not informed consent.
Treatment records need particular care. Records of substance use disorder treatment from federally assisted programs fall under their own federal confidentiality rule, 42 CFR Part 2, and the consent language for sharing them has to be written with that rule in mind. Have whoever handles privacy for the program review the form before it is used.
Where the program allows it, never make the service depend on agreeing to share data for evaluation. People who decline should be served the same way, and the program should report what share of participants declined, because a high rate tells you something about trust.
Avoiding perverse incentives
Every outcome measure changes behavior once money or reputation rides on it. Plan for that before you choose the measures, not after.
The most common problem is selection. A program paid on employment outcomes has a reason to enroll the people most likely to find work anyway and to quietly turn away people with the greatest needs. Report outcomes by the risk and need levels participants start with, and track enrollment against the population referred, so a program that selects easier cases shows up clearly.
The second is definitional drift. "Employed" becomes one shift of day labor. "Housed" becomes a couch for a week. Tight definitions in the table above exist to prevent this, and a sample of case files reviewed each quarter keeps them honest.
The third is subtler. A program measured on supervision compliance can drift toward reporting every lapse to protect itself, which raises violations rather than lowering them. Supervision outcomes belong on the dashboard as context, but they should never be the measure a partner is paid on.
An outcome measure that rewards a program for choosing easier people has stopped measuring the program and started measuring its intake.
Small numbers and privacy
Reentry programs often report by county, by partner or by demographic group, and many of those cells hold a handful of people. In a small county, "two participants returned to custody this quarter" may identify exactly who they are. Set a minimum cell size before publishing anything, and suppress or combine cells below it. Federal agencies set such thresholds for their own data; the cell size suppression policy CMS applies to the data it releases, for example, bars publishing counts from 1 to 10.
Suppression has to be done consistently. If a table shows a total and every cell but one, the hidden cell can be computed by subtraction, so suppress a second cell or report the total at a broader level. Combine small counties into regions, combine quarters into a year, and say on the report that small numbers were suppressed and why. Internal dashboards used by program staff under a data agreement can show more detail than anything published, but the access rules should be written down.
A dashboard a program officer can act on
A program officer has a few minutes and one question: is this working, and where is it not? The dashboard answers that and points to what to do next.
- One row per outcome, showing the current cohort, the prior cohort and the program's target
- Enrolled and completer counts shown together, with enrolled first
- Results by partner and region, with small cells suppressed or combined
- Results by starting risk and need level, so selection effects are visible
- Interim results labeled as interim for cohorts still inside their window
- The data source and its lag noted beside each outcome
- A short list of partners or areas off target, each with the next step and who owns it
- The consent decline rate, as a signal of how far participants trust the program
Resist adding more. A dashboard with forty measures gets opened once. A dashboard with seven outcomes, the context to read them, and a list of what to do next gets opened every month.
Measure with dignity
The people behind these numbers are returning citizens rebuilding their lives, often against long odds. The language a program uses about them shapes how staff, partners and funders treat them. Write "people returning home" or "participants," not labels that define a person by a conviction. Report what people achieved, not only what they avoided.
Ask participants what success looks like to them, and include at least one measure they would recognize as their own. A program that measures only what the system cares about will miss what keeps a person going: a child who calls back, a boss who trusts them with a key, a place where they belong. Those are the outcomes that last.
Frequently asked questions
Yes. It remains the outcome most funders and policymakers ask about. Define it precisely, as rearrest, reconviction or reincarceration, with a fixed start date and window, and report it alongside the outcomes that show progress earlier, such as housing, employment and treatment engagement.
Many use state unemployment insurance wage records, measured in the second and fourth calendar quarters after release or enrollment, the same approach workforce programs commonly use. Because those records miss self-employment, informal work and out-of-state jobs, programs supplement them with pay stubs or employer verification.
Check at 30, 90 and 180 days, and define stable housing as a place the person can stay for at least the next 90 days. A single check at exit misses the people who are housed in the first month and lose it by the third, which is when many housing arrangements fail.
Set a minimum cell size and suppress or combine anything below it, apply complementary suppression so hidden cells cannot be recovered by subtraction, and aggregate small areas into regions or short periods into longer ones. Keep detailed data in internal dashboards governed by written access rules.