Methods

This page is our transparency pledge. It explains how EDORA assembles, normalizes, and interprets data for the Learning Center, dashboards, state briefs, and case studies.

Data Vintages & Provenance

Timeframe: core series 2015–2025, with recency emphasis in dashboards that indicate it.
Provenance: State agency reports, court admin data, OJJDP, HUD AHAR, CDC/KFF, SAMHSA, provider evaluations (see Sources).
Documentation: Field definitions live in Data Dictionary; copy-ready TSVs in the Data Appendix.

Normalization & Denominators

Youth population rates: most justice metrics normalized to ages 12–17.
School metrics: normalized to enrolled students; specify year and enrollment basis.
Clear denominators: each page states the denominator; dashboards include tooltips where relevant.

Small-n Pooling & Flags

Sparse counts create noisy rates. We pool 2–3 years where necessary and set pooled_small_n=true. Pages and charts annotate pooled values to prevent over-interpretation.

Disaggregation & Equity

Where data permit, we disaggregate by race/ethnicity, disability, gender, and rural vs. urban.
Access and outcomes are shown side-by-side to surface disparity patterns, not just averages.

Breaks in Series & Comparability

When definitions or systems change, we mark a break in series and avoid stitching incompatible trends. Notes appear on the page and in dashboard tooltips.

Composite Index: YARI+

Domains: education exclusion, detention/diversion, SUD/MH proxies, foster care, housing proxies, violence.
Standardization: domain metrics are standardized before aggregation (e.g., z-like scaling or 0–1 normalization—documented per dashboard).
Weights: domain weights are explicit; recency weights bias 2019–2025 more heavily (see example in YARI+ (AR)).
Missingness: if a domain is missing for a county-year, we either (a) impute within documented bounds with a flag, or (b) re-weight available domains—policy noted on the dashboard.

Outcomes & Follow-Up Windows

Initiation: enrolled/admitted to service.
Completion: program finished as planned.
Improvement: short-run markers (attendance/behavior, credits, service completion).
Follow-up: 30/90/180/365-day recurrence or persistence indicators, when feasible.
Caveat: Diversion/reentry outcomes can be selection-sensitive; we document eligibility rules and overrides.

Screening & Risk Tools

Intake screens (e.g., risk/needs, MH/SUD) are for triage, not punishment. We distinguish public safety risk from service need and treat overrides as auditable events. See Screening & Risk Tools.

Reproducibility & Versioning

Schemas: canonical fields in Data Dictionary.
Exports: use semantic versions in filenames (e.g., county_panel.v1.tsv).
Per-page notes: each article/state brief lists sources and data vintage; dashboards display vintage and flags.

Design & Accessibility

Charts label denominators, vintages, and pooling; legends avoid color-only cues when possible.
All figures have descriptive alt text; tables provide screen-reader friendly headers.

Where to Go Next

Transparency note: pooled or proxy values are flagged on page; breaks in series are annotated; equity cuts are shown where data quality allows. Questions or reproducibility requests? Reach us via Contact.