Strike Panel¶
Comprehensive strike data covering South African industrial action during the late apartheid period.
Key Files¶
| File | Rows | Description |
|---|---|---|
strikes_master_csv.csv |
2,679 | Master strike incident panel (1984-1990) |
strikes_locations.csv |
2,678 | Location/geographic crosswalk |
strikes_with_macro_controls.csv |
— | Strike data with macro controls |
working_papers_converted/ |
31 files | Digitized strike working papers (WP 33-71) |
working_papers_extracted/ |
~480 files | OCR-extracted table fragments |
wip_extracted/ |
expanding | AI extraction from WPs 3-32 (1978-1984) |
Coverage¶
- Period: 1978-1990 (expanding backward to 1978 with WP 3-32 extraction)
- Incidents: 2,679 strikes in master panel (1984-1990) + pre-1984 incidents being extracted
- Firm matching: 222 strikes matched to JSE-listed firms with sufficient stock data for the per-strike regression specification used in the Naidu, Turban & Wilse-Samson paper
Pre-1984 Extension (2026-03-29)¶
wip_extracted/ — Claude API extraction from 25 SALDRU "Work In Progress" PDFs (WPs 3-32, Jan 1978 - Jul 1984). Uses text extraction where OCR layer exists, Claude Vision for image-only pages.
Output format: Company, Area, Region, Union, Issue, Workers, Date, WP_Number, WP_Date, Page
Early WPs (1978-79) are primarily editorial with few structured strike entries. WPs 21+ (1982-84) have consistent Company/Date/Workers/Union labeling yielding 20-40 incidents per issue.
Script: ops_admin/scripts/digitization/extract_strike_wps.py
Variables¶
- Date, duration, number of workers involved
- Firm name, industry, location
- Outcome (settlement, dismissal, etc.)
- Geographic identifiers for crosswalking
Key Findings (Naidu, Turban & Wilse-Samson, "Diffuse Pressure")¶
Three robust findings emerge from the matched 222-strike sample:
-
Aggregate Rand value loss: R 606 million (~$259 million USD) in announcement-day market value across the 157 strikes for which firm market values are available. The 1990 transition year alone accounts for R 271 million (45% of the cumulative total), coinciding with F.W. de Klerk's unbanning of the African National Congress and the release of Nelson Mandela.
-
Targeting null: Struck and non-struck JSE-listed firms had virtually identical Afrikaans board representation in 1983 (18.9% vs 18.8%, p = 0.94). National Party connections and board interlock centrality measures yield similar nulls. Strikes operated as a diffuse pressure across the formal economy rather than as a precision instrument targeting regime allies.
-
Temporal pattern (robust to specification): Strikes during the 1985-1987 State of Emergency produced effects indistinguishable from zero (mean AR ≈ −0.03%, p = 0.83), while strikes outside the Emergency, particularly in 1990, produced significantly more negative reactions. Consistent with strikes operating as information events whose marginal value depends on prevailing political uncertainty.
Individual strike effects are reported as a range across specifications:
| Specification | Mean AR | p-value | N |
|---|---|---|---|
| Dimson-corrected (primary) | −0.156% | 0.078 | 253 |
| Canonical (no Dimson) | −0.235% | 0.011 | 222 |
| Standard MacKinlay event study | −0.120% | 0.39 | 219 |
The Dimson-corrected estimate is reported as the primary specification because thin trading on the 1980s JSE is empirically present in the data (the Dimson-summed market beta is 1.099× the contemporaneous beta). Per-strike correlation between specifications is 0.51–0.80.
The mining-sector NUM result (canonical AR −0.44%, p = 0.024 under independence) is sensitive to controlling-shareholder clustering — most NUM strikes hit Anglo American gold subsidiaries, and clustering at the Big-Six controlling-shareholder level raises the p-value to 0.095. The paper reports this descriptively rather than as a clean causal estimate.
Methodology Note¶
The paper uses per-strike OLS regressions on a 3-day-aggregated event-time panel within [t_0 − 240, t_0 + 30] calendar days, not a textbook MacKinlay (1997) event study. The original 2013 producer code (Sebastien Turban) was located in May 2026 and faithfully reproduced in Python at correlation 0.9964 with the canonical AR file. Both implementations are provided in the paper's replication archive.
Provenance¶
- Source: Government records, academic working papers, Work in Progress (SALDRU)
- Compilation: Hand-coded from multiple sources (1970-1990)
- Source documents:
raw_archives/05_labor/strikes/