DNA Notification at Release in Pennsylvania State Prisons

The intervention is a standardized DNA notification delivered during the prison release process. The control group receives the usual release process with no DNA notification. Treatment 1 receives a brief spoken notice that the person's DNA profile has been entered into CODIS if the person was convicted of a felony or other qualifying offense, plus a narrow FAQ. Treatment 2 receives a longer spoken notice that includes the same core CODIS reminder plus additional information about how DNA is used in investigations, along with a broader FAQ.

2026-06-01

2027-06-01

Any rearrest within T* months of release.

The primary outcome is a binary indicator equal to one if an individual is rearrested within T* months of release and zero otherwise. T* will be selected before treatment effects are estimated using a pre-specified elbow rule applied to cumulative rearrest rates over months 1 through 24 in a historical pre-intervention cohort. If a sufficiently comparable historical cohort is not available, the same rule will be applied to the study control group only before outcome analysis. T* is defined as the first month at which the month-to-month increase in cumulative rearrest probability falls below 10 percent of the maximum monthly increase for three consecutive months. If no such month occurs by month 24, T* will be set to 24.

Any rearrest within 3 months of release.
Any rearrest within 6 months of release.
Any rearrest within 9 months of release.
Any rearrest within 12 months of release.
Any reconviction within 3, 6, 9, 12, and T* months of release, if available.
Return to incarceration and or technical violation outcomes within 3, 6, 9, 12, and T* months of release, if available.
Time to first rearrest from release.
Crime-type-specific post-release outcomes, using PA DOC offense classifications.

Most secondary outcomes are direct administrative indicators measured at fixed post-release horizons. Time to first rearrest will be measured from the release date to the first rearrest date, with right-censoring at the end of the observation window. Crime-type-specific outcomes will use PA DOC's offense classification system. The main category will be used for the primary crime-type families and, where sample size permits, subcategories will be used for more granular exploratory analyses. If categories are too fine-grained for power or interpretability, they will be coarsened into a smaller set of mutually exclusive categories such as violent, property, drug, and public order or other before outcome analysis.

This is an individual-level randomized controlled trial embedded in the Pennsylvania Department of Corrections release process. Individuals released from participating state correctional institutions are assigned to one of three study groups during the exit interview workflow. The control group receives the usual release process with no DNA notification. Treatment 1 receives a brief DNA notification and narrow FAQ. Treatment 2 receives a fuller DNA notification and broader FAQ. Outcomes will be measured using administrative records after release.

Not available

Administrative deterministic assignment based on the second-to-last digit of the incarcerated person's PA DOC identification number. The mapping is pre-specified: 0 to 3 Control, 4 to 6 T1, 7 to 9 T2. Staff do not choose assignment.

Individual release event.

No

Not clustered. Approximately 21 state correctional institutions are expected to contribute observations, but treatment is randomized within facilities at the individual release-event level.

Approximately 10,331 individual release events over one year before study-specific exclusions.

Approximately 4,132 control release events, 3,099 T1 release events, and 3,099 T2 release events, assuming one year of rollout before study-specific exclusions.

The main outcome is a binary indicator for any rearrest by T* months. Using PA DOC's 2025 facility release counts as the planning benchmark, a one-year rollout yields approximately 10,331 release events before exclusions. Under the planned 40/30/30 allocation, the expected group sizes are about 4,132 control, 3,099 T1, and 3,099 T2. At a benchmark control-group rearrest rate of 0.35, the binary-outcome standard deviation is sqrt(0.35 x 0.65) = 0.477. Under the exact two-sample difference-in-proportions calculation with alpha = 0.05, two-sided, and 80 percent power, the minimum detectable effect is 3.1 percentage points for T1 versus Control and T2 versus Control, and 3.4 percentage points for T1 versus T2. Under a conservative 75 percent delivery scenario, the treatment-versus-control MDE rises to about 4.2 to 4.4 percentage points. As a clustering sensitivity, simulation-based power calculations that allow facility-level ICC up to 0.01 yield MDEs of about 3.3 percentage points for treatment versus control and about 3.5 percentage points for T1 versus T2, reflecting individual randomization within facility and analysis models with facility fixed effects.

Pre Analysis Plan