Skip to content

Value Added (va_consolidated)

The consolidated Stata pipeline behind the Common Core Value-Added paper (Carrell, Kurlaender, Martorell, Naven, Sun — "Do Schools Matter?"). It cleans the raw data, builds the estimation samples, runs the value-added estimation, and produces every table and figure in the paper.

This is the most fully documented repo in the lab and the best model to read first — it ships a plain-English HANDOFF.md written for someone comfortable with Stata but new to git and Claude Code.
GitHub https://github.com/chesun/va_consolidated
Runs on (Scribe) /home/research/ca_ed_lab/projects/common_core_va/consolidated
Entry point do/main.do
Status Offboarding deliverable — acceptance run in progress
Start with HANDOFF.md

What it is

School-level value-added estimates for ~1,400 California high schools across four cohorts of 11th-graders (2014–15 through 2017–18), using a CFR-style "value-added with drift" estimator, plus a survey-VA mechanism analysis linking the estimates to CalSCHLS school-climate indices. It is the consolidated successor to two predecessor codebases (cde_va_project_fork and caschls).

How to run it

On Scribe, from the project folder:

cd /home/research/ca_ed_lab/projects/common_core_va/consolidated
nohup stata-mp -b do do/main.do &      # full run ~36–38 hours — plan for two days

The run executes seven phases in dependency order (data prep → samples → VA estimation → VA tables → survey-VA → paper outputs → automated data checks). VA estimation (Phase 3) is the bottleneck and dominates the ~36–38-hour wall-clock total — plan for two calendar days. Phase toggles near the top of do/main.do let you re-run just part of it. See the HANDOFF for partial-run rules and the m4_acceptance_run switch.

Documentation map

  • HANDOFF.md — plain-English run & reproduce guide (start here).
  • README.md — fuller technical reference.
  • decisions/ — numbered ADRs explaining why each load-bearing choice was made.
  • quality_reports/ — review reports and the golden-master reproduction evidence.

Who to ask

  • Christina Sun — author (christinasun101@gmail.com / ucsun@ucdavis.edu).
  • Paper LaTeX is a separate repo (va_paper_clone) — the pipeline produces the tables/figures the paper consumes.