Replication & Reproducibility¶
A book and a paper draw on the datasets catalogued here. This page says where each one lives and how its results are reproduced.
The repositories¶
The project is split across four private Git repositories. This archive is the data layer that the other three read from.
| Repository | Contents | How results are reproduced |
|---|---|---|
sa-apartheid-economy-data (this site) |
Datasets, corrections ledgers, panel builders, and this documentation | The builders in shared_resources/scripts/panel_builders/ rebuild the panels from the source files and the ledgers |
apartheid-labour-shock |
Boone and Wilse-Samson, Coalition Fracture Under Labor Scarcity (the 1974-shock paper) | paper/replication/run_all.py rebuilds every reported number and checks the headline coefficients |
apartheid-book |
The Cost of Apartheid manuscript and its analysis code | replication/MANIFEST.md maps each table and figure to its script and data |
strikes-apartheid-jpe |
The rebuilt strike file and the strike and sanctions tests (with Naidu and Turban) | build/run_all.py rebuilds the evidence used in book Chapters 5, 8 and 10 and checks the headline numbers |
Main results¶
| Result | Value | Data |
|---|---|---|
| Eiselen electoral divergence | East × 1987 = +14.2 pp right-wing vote, 172 divisions, N = 434 | Election returns and the date-correct crosswalks |
| National wage shock | Black farm wages roughly doubled after 1974; open and closed districts adjusted alike (open × post = 0.018, p = 0.47) | Agricultural census panel |
| National mechanization, 1972–80 | Combines +30%, tractors +13%, regular Black employment −10% | Agricultural census panel |
| Combines, open vs closed districts | Not robust: +0.017 (p = 0.84) with district and year fixed effects | Agricultural census panel |
The book's table of estimates (docs/canonical_estimates.md in the book repository) is kept in line with the paper. If a number here disagrees with the book, the book is right; please write to us.
How far the data were checked¶
See Provenance & Checks. In short:
- The 2012–13 contract transcriptions are unchecked.
- Internal arithmetic checks have been run on all years.
- The agricultural census for 1961–83 was corrected by hand by research assistants.
- For the cells the paper uses, the versions were compared, and every disagreement was read from page images by AI.
- Corrections are ledger rows that the builders apply.
DOI and versioning¶
The archive has no DOI yet. The paper's replication package will be deposited on Zenodo when the paper is accepted, and the book's on publication. Until then, cite this site and the date you consulted it. Restricted inputs, and how to request them, are described in Licensing & Access.