EDORA
Skip to content
∑

EDORA • Learn

Justice Index · Advocacy Lab · Field Guide

EDORA Learn — Methods

← Back to Learning Center

Follow-Up Windows and Cohort Construction

Follow-up windows define how long a study observes outcomes after a key event such as release, program entry, or case closure. Cohort construction defines who enters the study and when their observation begins. Both choices strongly influence reported rates.

Defining a Follow-Up Window

A follow-up window is the time span in which outcomes are measured after a reference date. Common intervals are 6, 12, and 24 months. Shorter windows capture early outcomes but miss long-term patterns; longer windows reveal persistence but suffer more attrition.

  • Fixed windows: Exactly 180, 365, or 730 days after a defined start event.
  • Rolling windows: Moving observation frames that update continuously with each new participant.
  • Administrative windows: Aligned with calendar or fiscal years, often for reporting convenience.

Constructing Cohorts

A cohort is a group of individuals who share a common starting event and are tracked over time. Defining cohorts consistently ensures that each observation has the same exposure opportunity. The most comparable studies use closed cohorts—those that admit no new cases after a set start period.

  1. Identify the start event: release, admission, or program start.
  2. Set a fixed entry window: for example, “all youth released in 2021.”
  3. Observe for a set duration: 6, 12, or 24 months after the start date.
  4. Mark censoring: note any early exits, transfers, or missing data.

Why the Choice Matters

  • Short vs. long windows: Short windows emphasize immediate outcomes but may undercount delayed reoffenses. Long windows improve completeness but require more maintenance and produce lagged reporting.
  • Open vs. closed cohorts: Open cohorts continuously add participants, mixing exposure times and complicating annual rates.
  • Reporting cycles: Aligning windows with reporting years may improve usability but sacrifice precision in true elapsed time.

Data & Methods

The research file notes that most justice and human service datasets record event-level timestamps for intake, release, or completion. Follow-up windows should be calculated directly from those timestamps, not from report-year aggregates. Cohort boundaries should be stated in each visualization or table. When comparing systems, readers should verify whether the same window and cohort logic were applied.

Related

Transparency note: When comparing rates, always check whether each system uses fixed-day follow-up windows and closed cohorts. Differences in these assumptions explain many apparent gaps between datasets.