
Replication archive for "Load Shedding and Crime in South
Africa: Causal Estimates, Mechanism Evidence, and the
Distributional Welfare of the Energy-Apartheid Externality."
Contents:
- PAPER/ — manuscript source (Overleaf .tex), rendered PDF
- CODE/ — analysis scripts: data ingestion, paper regressions,
Session 22-28 additions on the scheduled-vs-unscheduled
question, and shared helpers
- DATA/ — parquet snapshots of every database table the paper
uses (~280 MB across 34 tables), schema DDL, station
crosswalks, third-party-source provenance
- RESULTS/ — per-session output directories (tables and figures)
- DOCUMENTATION/ — 33 session reports + methodological notes +
section drafts
All data sources are public and properly cited. Restoration to
a fresh Postgres database is documented in DATA/README.md.
The headline causal estimates (a −1.26% decline in residential
burglary and a −1.23% decline in theft from motor vehicle per
+100 outage hours, both p<0.0001 on 200-iteration bootstrap)
are reproducible from DATA/panels/ via the scripts in
CODE/analysis/. Seven indirect tests of the
behavioural-adaptation channel (three unscheduled-outage proxies
in §5.8 plus four reframed identification strategies in §5.9)
collectively reject adaptation as a population-level mechanism
while identifying a category-specific channel for non-residential
burglary alone.
Built: 2026-05-11 against v5.2 manuscript.