Find What Works: Schools
← All source documentsRaw file
Source document

docs/analysis/absenteeism/bakeoff/README.md

Served verbatim from the project repository. Internal working document conventions apply: documents may reference file paths, branch names, and findings-ledger anchors from the repo.

Chronic-absenteeism method bake-off — results & review package

Companion to the pre-registered design memo (../03_bakeoff_design.md). The bake-off selected candidate (c), the WLS residual-z, as the chronic-absenteeism status measure, with a within-year-z trend (direction + magnitude over 2022-23 → 2024-25, plus a dual-endpoint persistence gate). Full recommendation, criteria, and the six open editorial decisions are in the memo.

Views — served from public/bakeoff/

Self-contained HTML (inline CSS/JS, no build step). Reachable at /<file> on the site once this is merged; on this branch they appear only in the Vercel preview. Every page is noindex, nofollow, internal debug links are stripped, and names are leads to investigate, not verdicts (the validation gate applies to an advertised/published story, not this working draft).

  • 04_bakeoff_results_review.html — results memo (start here)
  • 02_descriptives_packet.html — descriptives background (overall trends, grade splits)
  • named_outliers_WIP.html — status outlier candidates
  • trend_candidates_WIP.html — trend candidates (z-path + actual %, persistence gate)
  • grade_control_sensitivity_WIP.html — grade-control (spec E vs F) sensitivity

A shared "Working views" nav links them together.

Source / reproducibility — this folder

  • 04_bakeoff_results.md — memo source (the memo HTML is rendered from it)
  • render_results_html.py, render_named_outliers_html.py, render_trend_html.py, render_spec_sensitivity_html.py — view generators
  • compute_trend.py → trend.csv, trend_meta.json — trend layer (within-year-z change 2022-23 → 2024-25 + the dual-endpoint persistence gate, race-unaware)
  • pooled_trend_diagnostic.py — decision-6 diagnostic (pooled + year-FE vs within-year; Δz rank-correlation 0.97, ~13/17 worsening & 13/14 improving names common)
  • export_public_views.py — regenerates the public-safe copies in public/bakeoff/ (strips localhost debug links, forces noindex)

Upstream compute (per-year candidate fits, the grade-mix index gmx, the criteria evaluation) lives in the non-git working tree, not the repo.

For review

The open decisions are in the memo's "Candidate schools & decisions for the checkpoint" section: (1) race in the model, (2) D11 strict vs loose composition-stability, (3) naming policy + validation gate, (4) charter coverage, (5) trend framing, (6) within-year vs pooled trend expectation. This is a review request — not published to the production site (that happens on merge).