Geographic Crosswalks¶
Magisterial districts were renamed, split and merged over time, and different agencies coded them differently; the electoral divisions of the House of Assembly were redrawn by a Delimitation Commission about every eight years. This page describes how the archive links these units, and the two district classifications used in the analysis.
Magisterial districts¶
The common geography is the roster panels/master_district_roster.csv: 279 districts on the 1960/61 boundaries, 67 of them in the homelands.
File (in shared_resources/data/) |
Links |
|---|---|
district_crosswalk_master.csv |
Pre-1949 agricultural census names to roster districts. Wrong matches are rejected through a ledger |
teba_panel/station_district_crosswalk_v4.csv |
TEBA recruiting stations to districts (research use only; see Licensing & Access) |
Districts with the same name in different provinces (Heidelberg, Middelburg, Richmond) are matched by province, never by bare name. The matcher in panel_builders/district_names.py refuses to guess an ambiguous name.
Electoral divisions to districts¶
Each election is described by the delimitation in force when it was fought. A division name can cover different ground under different delimitations, so divisions are never linked across delimitations by name. For example, the 13th Delimitation abolished the old De Aar division and renamed Colesberg as De Aar.
| Elections | Delimitation | Crosswalk | Weights |
|---|---|---|---|
| 1966, 1970 (and, as an approximation, 1948–61) | 12th (1965/66) | crosswalk_1966.csv: 160 divisions |
Estimated voters on the 31 May 1973 roll |
| 1974, 1977 | 13th (1974) | crosswalk_1974.csv: 165 divisions |
Estimated voters on the 31 May 1973 roll |
| 1981, 1987 | 14th (1980) | crosswalk_1980_integrated.csv: 165 divisions |
1960 white population, or area |
1980: from the delimitation maps¶
The 1980 crosswalk has one row for each division–district pair, with area and population shares and a source and confidence for each pair. Four sources are combined, in this order of precedence:
- The delimitation maps, read by computer vision. AI agents (Claude) georeferenced the 67 rural 1:250 000 sheets and traced the red division boundaries, giving 61 rural divisions as polygons; 42 sheet overlays passed the agent's own check, 23 were marked for checking and 3 failed. No person has reviewed the overlays.
- A research assistant's map check. Boone's research assistant checked a Gemini list of pairs against the same maps; the vision reading recovers 198 of the 212 pairs in that set.
- The 14th Commission report. Two separate AI readings agree on every transfer. Where the text contradicts the map reading, the text is used (Ngotshe goes to Umfolozi).
- A Gemini reading of the report, used only where no other source covers a pair.
By division:
| Basis | Divisions |
|---|---|
| Map reading only | 50 |
| Map reading plus the research assistant's check or the report text | 11 |
| Research assistant's check or report text, without the map | 12 |
| Gemini reading only | 92 |
Of the 92 Gemini-only divisions, 61 are urban seats "within the X district", where the district is not in doubt; 28 are Rand, Vaal and Free State divisions on missing map sheets, all east of the Eiselen Line; and 3 are Cape Peninsula seats, all west of it. Of the 212 roster districts outside the homelands, 209 are fully allocated; the exceptions are Chatsworth, which no division claims, and Maclear and Mafeking, whose remainders lie in the Transkei and Bophuthatswana. Walvis Bay has no South African district.
1974 and 1966: chained back from the reports¶
These two crosswalks were not read from maps. Each starts from the 1980 composition and undoes, in reverse order, the boundary changes stated in the later commission report: 205 transfers in the 14th report and 221 in the 13th. Two AI agents read each report separately, except pages 15 and 45–49 of the 13th report (Peninsula and Witwatersrand urban detail), which only one read. Where a report gives no voter count for a transfer, the count is solved from the report's arithmetic. Against the voter totals the reports state, 59 of 62 divisions match within 10 voters for 1974 and 56 of 72 for 1966.
The shares are estimates. Which districts lie in which division is the firmer result, and each row carries a confidence value.
Limits¶
- Several map sheets are missing or show no division lines over parts of the Free State and East Griqualand; districts there are allocated from the lower-ranked sources. The metropolitan insets were not processed, so the urban seats rest on the report text and the Gemini reading.
- In the 1974 and 1966 crosswalks, transfers within the big metropolitan areas were assigned to districts automatically, and their split between neighbouring metropolitan districts has low confidence.
- The older files
crosswalk_electoral_magisterial_districts_1980_clean.csv,crosswalk_electoral_magisterial_1974.csvandcrosswalk_electoral_magisterial_districts_1966.csvare LLM readings of the report narratives and are incomplete. Do not use them.
Division traits¶
Each crosswalk gives the share of a division lying in districts that are east of the Eiselen Line, open to recruiting, closed, high-potential, or in a homeland (division_traits_1980_integrated.csv, and division_traits_by_year.csv in the 1974-shock package). The main specification uses majority dummies; the robustness checks use the shares. Under any weighting, no division changes side of the Eiselen Line compared with a single-district coding.
Open and closed districts¶
The classification of districts as open or closed to mine recruiting comes from a 1974 NRC board memorandum (NRC File B1061 Pad 2). That classification is reproduced as Figure 1 of Crush (1993), "The Long-Averted Clash", and was digitized from the figure (open_close_70_temp.xls). The categories are:
black_whole: open to recruiting of Black labour;black_part: partly open;grey: closed, but rated by the NRC as having high recruiting potential;white: closed (white farming areas).
The Eiselen Line¶
The Eiselen Line is a separate policy from the open/closed classification. W. M. Eiselen, Secretary for Native Affairs, defined it in 1955 as the eastern edge of the Coloured Labour Preference Area: "South of the Orange River and West of the Magisterial districts of Gordonia, Hope Town, De Aar, Hanover, Richmond, Murraysburg, Aberdeen, Willowmore, Uniondale and Knysna" (Goldin 1984). West of the line, farms relied mainly on Coloured workers. East of it, they relied on Black workers drawn from the same pool the mines recruited from. The roster's eiselen_line column holds the coding.