EDORA
Skip to content
∑

EDORA • Learn

Justice Index ¡ Advocacy Lab ¡ Field Guide

EDORA Learn — Methods

← Back to Learning Center

Recidivism: Definitions, Windows, and Comparability

The research source compiles many definitions of “recidivism.” This page distills the common variants so readers can understand who is counted, what outcomes qualify, and how follow-up windows alter the rates reported.

What We Track

  • Outcome definition: Many series count a new delinquency adjudication or criminal conviction. Others use arrests, petitions, filings, or returns to supervision. Technical rule violations may or may not be included—always check the footnotes.
  • Windows: 6-month, 12-month, and 24-month follow-ups are most common. Some sources use fixed-day windows (e.g., 365 days); others use “by end of next year” or “until supervision closes.”
  • Cohorts: Closed cohorts by release year (post-custody) or start year (diversion/probation) are more comparable than rolling denominators.
  • Unit of measure: Person-based rates (one youth counted once per window) vs. event-based counts (multiple outcomes per person). Person rates are easier to compare.
  • Exclusions: Status-only findings and technical violations are sometimes excluded from “new offense” measures; where they’re included, labels should make that explicit.

Typical Flow

  1. Define the cohort (e.g., all youth released in calendar year 2021).
  2. Fix the window (e.g., exactly 365 days for the 12-month rate).
  3. Link outcomes to each person within the window and keep the first qualifying event.
  4. Compute the rate: people with a qualifying outcome á people in the cohort.
  5. Document caveats: definition differences, series breaks, pooling, or suppressions.

Comparability & Caveats

  • Definition drift: Arrest-based measures typically run higher than adjudication/conviction measures. Mixing them obscures trends.
  • Window alignment: A “12-month” label can mean different things. Prefer fixed-day windows when available; otherwise, note the convention used.
  • Program context: Post-custody, probation, and diversion cohorts are not interchangeable. Compare within the same cohort type.
  • Small-n instability: Tiny cohorts swing widely. Many sources pool years or suppress cells; check notes before interpreting outliers.

Data & Methods

This summary is synthesized from the provided research text. Where multiple definitions appear, parallel series can be constructed (e.g., adjudication-only vs. arrest-based) as long as the underlying fields exist. Follow-up windows should be derived from recorded exit/start dates; any deviation (such as “by next calendar year”) should be labeled directly in charts and tables.

Related

Transparency note: Where cohorts are small, pooling of adjacent years and clear series-break labels help avoid false signals and protect privacy.