Census, race classification and demographic data¶
Did censuses neutrally describe or actively construct racial identities, and how reliable were apartheid demographic data?
Works engaging the debate: 6
Positions¶
- Constructivist: the state creates categories (Scott, Kertzer and Arel)
- Race has no scientific basis
- Apartheid-era projections and censuses distorted for political ends
Evidence base¶
The debate rests overwhelmingly on the published output of the census itself. The two Christopher studies, each recorded twice in the evidence ledgers (Christopher (2002a)/(Christopher (2002b)) and Christopher (2009a)/(Christopher (2009b))), read in sequence the printed reports, questionnaires, introductory statements and published tables of nearly forty enumerations. These run from the Cape census of 1865 (G20-'66) and its successors of 1875, 1891 and 1904, through the Natal (1891, 1904), British Bechuanaland (1891), Orange Free State (1880, 1890), South African Republic (1890) and 1904 Transvaal and Orange River Colony censuses, to the Union counts of 1911–1960 and the Republic counts of 1970–1991. They also cover the Transkei, Bophuthatswana, Venda and Ciskei bantustan censuses of 1985–1991 and the Statistics South Africa censuses of 1996 and 2001, together with the 2007 Community Survey. The unit is the national or colonial census report, and the quantitative content is descriptive. In Christopher (2002a) this takes the form of a Table 1 of classification categories, 1865–2001. In Christopher (2009a) it consists of summary tables of coverage, White-only counts (1918, 1926, 1931, 1941), nationality tabulations (1921, 1926, 1936) and citizenship returns (1980, 1991). Reports are set against legislation: the ZAR law compilation of Jeppe & Kotze, the Cape Government Gazette notice of 1847 on Native Locations, the South Africa Act, the Population Registration Act of 1950 and the Employment Equity Act of 1998. Three Cape Town deeds of transfer (1924, 1949, 1951), the Chinese Association v Minister of Labour judgment (2008) and a 2005 UN Special Rapporteur report supply illustrative material.
Two works use census data as demographic evidence rather than as text. Simkins and Van Heyningen (1989) applies indirect estimation to the age, sex, marital and birthplace tables of the 1891 and 1904 Cape censuses. The methods are Preston-Bennett, back-projection, and stable-population and Coale-Demeny model tables. The census results are cross-checked against the 1895–1903 reports under the Births and Deaths Registration Act, the Cape Statistical Registers and Labour Commission wage evidence. From this the authors estimate fertility, mortality and migration for six social and ethnic groups. They link these estimates to Union and Republic tabulations from 1911 to 1980, unpublished 1980 census raw data, and the Sadie and Grobbelaar projections. Zuberi et al. (2005) works at the individual and household level. Its main sources are the 10% and 30% microdata samples of the 1996 census, the full-count 1991 microdata with Central Statistical Services weights and HSRC aerial-survey adjustments, and the October Household Surveys of 1995–98. To these it adds Population Register death data, a 15% sample of death notification microfilm, and official life tables for 1921–1996. It is the only work to draw on state archive files, press coverage of the 1991 census and an interview with a UNFPA census adviser.
Contested evidence¶
The central interpretive split concerns what the persistence of the four-fold classification shows. Christopher (2002a) and Christopher (2002b) treat the 1996 finding that 99.1 per cent of respondents placed themselves in apartheid-era categories as evidence that the census 'locks in' the groups it observes. That inference is supported only by Barrier's and Martin's theory, not by evidence on how respondents understood the question. The same figure could equally be read as a response to lived legal and social realities, or to the absence of alternative answer options. Zuberi et al. (2005) and Simkins and Van Heyningen (1989), by contrast, use race as a stable analytic variable for measuring inequality in life expectancy, infant mortality, schooling and household structure. In doing so they implicitly accept the categories the constructivist position treats as artefacts.
This creates a tension of comparability. Christopher documents that the Cape taxonomy was revised between 1875 and 1904 and simplified after 1911. In 1970 'Hottentots' were transferred to Coloured, and Christopher infers, without demonstrating it, that this deflated the African count in the western Cape. Yet (Simkins and Van Heyningen (1989)) builds intercensal estimates on group definitions held constant between 1891 and 1904, and links them to a long-run life-expectancy series running to 1946 and beyond.
On reliability, the works rest on partly incompatible foundations. Christopher (2009a) reads the 1980 citizenship returns, in which Whites outnumbered African South African citizens, and the 1991 result of 93.9 per cent African South African citizenship as evidence of political design and its failure. However, it explicitly lacks any evidence on undercount. Zuberi et al. (2005) shows that the 1991 count badly undercounted Africans and depended on HSRC inflation. It further shows that apartheid-era projections overestimated African fertility, and that pre-1985 life tables assumed complete death registration and so understated mortality. Simkins and Van Heyningen (1989), meanwhile, uses the Sadie and Grobbelaar projections and the 1980 census as benchmarks for its long-run conclusions. The 'distortion' position is thus documented demographically by Zuberi but only textually by Christopher, and the two kinds of evidence are never directly reconciled.
Underused and missing sources¶
Every evidence ledger notes the same archival gap. None of the works uses enumerator instructions in a systematic way, nor the internal files of census directors and the Office of Census and Statistics, nor Population Registration Board reclassification records, nor parliamentary debates on census design. Zuberi et al. (2005) touches state archives and press coverage only selectively. Post-enumeration and undercount studies by group, cause-of-death registration by race and district, and surveys of self-identification are also absent. These are precisely the sources that would test whether categories shaped identity or merely recorded it, and whether apartheid figures were biased by design or by capacity. The bantustan censuses (Transkei, Bophuthatswana, Venda, Ciskei) appear in only two works and are described rather than analysed. Vital registration microdata for Africans before 1990, which would let fertility and mortality estimates be checked against independent registration, are wholly missing.
The census microdata series for 1960–2011 permits a district-level test of Christopher's claim about the 1970 Hottentot reclassification. It also allows cohort comparisons of the 1991 and 1996 counts to quantify undercount by race, age and sex, and a check on the stability of self-classification from 1996 to 2011. The population distribution tables for 1904–1960, together with the 1960 and 1970 tables, cover the period of White-only and race-specific Union enumerations. They could be used to assess how much non-uniform coverage after 1911 disrupts long-run series. The 2011 small-area shapefiles allow spatial analysis of post-apartheid classification.
Strength of key claims¶
| Basis | Share of key claims |
|---|---|
| Systematic data | 40% |
| Partial data | 44% |
| Illustrative or anecdotal | 9% |
| Asserted | 7% |
Comparative figures for all debates are on the Literature overview.
Best-evidenced works¶
Zuberi et al. (2005), Simkins and Van Heyningen (1989), Christopher (2009a), Christopher (2002a)
Most-used primary sources¶
Holdings are matched by machine against the archive's file manifest; a dash means no match was found, not that the project certainly lacks the source.
Works and their positions¶
| Work | Position |
|---|---|
| Christopher (2002a) | Sides with Barrier: census administrations create and reinforce the identity of the groups they were meant only to observe, and the South African case supports this, especially for the Coloured population. |
| Christopher (2002a) | Agrees with Suzman and colonial census officials that no scientific basis exists. Race definition is an administrative and political construct that shifted with state purposes. |
| Christopher (2002b) | Constructivist: sides with Barrier (1981) that census administrations create and reinforce the identities of the groups they were meant only to observe, and sees the Coloured case as supporting this. |
| Christopher (2002b) | Sceptical: quotes Suzman and the Malaya census commissioner to show that no uniform or scientific basis for race classification exists. |
| Christopher (2009a) | Argues that South Africa does not fit the pattern of colonial-to-post-colonial continuity. White minority rule from 1910 to 1994 broke the link, which was restored only nominally after 1994. |
| Christopher (2009a) | Sides with the constructivist view that the state's categories shape identity. The census projected each regime's political agenda, and race labels persisted across regimes. |
| Christopher (2009b) | Argues South Africa is an exception to the continuity noted elsewhere, because white minority rule from 1910 to 1994 broke the link; it was restored only nominally after 1994. |
| Christopher (2009b) | Follows Scott, Anderson and Kertzer & Arel: the census projects the state's national image and creates the categories it claims to describe. |
| Christopher (2009b) | Census evidence shows exclusion failed: the white share could not be disguised and 93.9% of Africans were returned as South African citizens by 1991. |
| Simkins and Van Heyningen (1989) | Argues the white fertility transition began c.1890, with Coloured and black fertility at natural levels. Racial inequality stretched the overall transition in a historically unprecedented way. |
| Zuberi et al. (2005) | Khalfani et al. treat racial classification as a state-built social construct that served stratification and distorted statistics. They nonetheless support retaining race categories post-1991 for redress and demographic analysis. |
| Zuberi et al. (2005) | Udjo argues African fertility in the 1970s-80s was overestimated, partly for political reasons, which explains the lower-than-projected 1996 census count. |
| Zuberi et al. (2005) | The editors argue that the 1991 census seriously undercounted Africans, and that demographic inflation of African males may have been shaped by political aims. They treat 1996 as the most complete enumeration. |
| Zuberi et al. (2005) | Bah finds estimates sensitive to the cause-of-death lists used, with HIV/AIDS the leading cause of death by the 2002 sample. Udjo says the evidence does not support attributing rising child mortality mainly to HIV. |