Back to contents index

During October 1996, under the Census ’96 motto ‘Count us in’, a hundred thousand temporary employees of Statistics South Africa (then the Central Statistical Service) fanned out across the cities, towns, townships, informal settlements, villages, farms and rural areas of the country. Their task was to record the details of people living in more than nine million households of South Africa, as well as those in hostels, hotels, prisons and other institutions.

Following a mandate from parliament, the enumerators had been recruited predominantly from the ranks of the unemployed, half-a-million of whom applied for the available jobs. The successful applicants were trained for three days, and then undertook the fieldwork: 86 000 enumerators supervised by 16 000 chief enumerators and 2 000 controllers, reporting to 72 regional offices. Using questionnaires translated into all 11 official languages, enumerators sought up to 50 items of information on each person present on the reference night of 9-10 October, and also more than a dozen items on the nature of the household.

During the preceding 12 months, Stats SA demarcators divided the entire country into some 86 000 ‘parcels’ of approximately 150 households each. These were called enumerator areas or EAs. About half of the EAs spanned the informal settlements, rural areas, homeland areas and former ‘TBVC’ (Transkei-Bophuthatswana-Venda-Ciskei) states, in which roughly half the population is to be found. Previously, in the 1991 census, the detailed life circumstances and aspirations of these people had been largely ignored. In 1991, some of the populations in these areas had been counted by means of a ‘sweep count’ of entire magisterial districts, whilst most had not been visited at all. Instead, they had been statistically estimated from selective aerial photography and samples, or through the use of a demographic model reaching back 20 years.

By contrast, Census ’96 – the first nationwide census since the splitting up of the country under apartheid after 1970 – sought to apply the same methodology to everyone: visiting the household, and obtaining details about all its members from a representative who was either interviewed, or else filled in the questionnaire, in the language of choice.

During the following month, the most skilled of the Stats SA enumerators were retained to re-visit 1% of the enumerator areas, list afresh the households there, and visit them again to ask who had not been reached, or reached elsewhere, during the census period. This exercise, called a ‘post-enumeration survey’ (PES), is conducted in order to estimate the extent of persons or households who are unavoidably missed in any census. This may be because they are difficult to reach, or refuse to be interviewed, or because some enumerators or supervisors are less thorough or efficient than others. The results from the PES are compared with those from the wider census by a process of ‘matching’, which yields an adjustment for ‘undercount’, i.e., the people who were missed.

In the case of Census ’96, Stats SA published some preliminary estimates of size of the South African population during mid-1997. The preliminary estimates were based on a manual tally of a sample of census questionnaires. They were adjusted for undercount by a simple application of the post-enumeration survey.

In the ensuing period, a total of 5 000 contract staff working over three shifts in nine provincial processing centres coded the data on the nine million completed census questionnaires, and entered this onto computers. During the same period, the laborious and thorough process of PES matching was undertaken, comparing results obtained in the post-enumeration survey with those obtained for the people in the corresponding 1% of EAs in the actual census. By these means, Stats SA has compiled the full set of data for publishing and further analysis by users, and final PES-based estimates have been obtained.

The final count for the population of South Africa as at October 1996, after PES-based adjustment for undercount, is shown in Table 1 below. By way of comparison, the preliminary estimates are also shown, as well the projections to 1996 of the totals which were derived from a demographic model in 1991.

*     All numbers given in this report are rounded to whole numbers. The totals may therefore differ
        slightly.

**    The percentages are rounded to the first decimal place, therefore they may not always add
        up to exactly 100.

Table 1 shows that the population of South Africa, according to the final count, after PES-based adjustment for undercount, is 40,6 million. This is some 1,5 million less than projections from the 1991 demographic model, and some 2,7 million higher than the preliminary estimates.

There are two reasons why the preliminary estimates were too low. Firstly, the simple analysis of the PES relied on the say-so of an informant in each household as to which members of the household had been reported upon or omitted during the actual census. The informant may have been a different person from the one who gave the original information during the census, and may not have known or reliably remembered what occurred. This was

why Stats SA subsequently undertook the process of actually seeking to match the answers on the questionnaires with the information about the corresponding household in the PES. The comparison revealed that informants tended to have under-reported those who had been missed. As a result, the final calculation of the undercount in the census rose to 10,7% from the 6,8% applied during the preliminary estimates.

Secondly, the preliminary estimates were based on a hand tally from a sample of the questionnaires as these were being readied for coding and capturing onto computers. Subsequently, it became clear that some of the provincial processing centres had not submitted all their administrative documents at the time that the sample was drawn. These questionnaires were, however, brought into account during the process of capturing them onto computers, and included in the improved undercount calculations – thereby increasing the final totals.

The right-most columns of Table 1 show that the proportions of men and women in the population are 48,1% and 51,9% respectively, according to the final estimates. The proportion of men is rather smaller than was previously expected by demographers, who applied adjustments so that it would be about 49,5% to 50,5%, as shown in the left-most columns, for the projections. However, one finds that the measured male proportion from Census ’96, after adjustment for undercount, corresponds very closely – in total and when broken down by population group – with that from the census in 1970. (The 1970 census was similar to Census ’96 in seeking to measure rather than merely model the attributes across the entire population. In both cases the number of foreigners present in South Africa – a large proportion of whom, especially in 1970, were male migrant workers – is removed before comparison.)

It also turns out that there is greater correspondence of the male-to-female proportion in the youngest age category, 0-1 years, between both the 1996 and the 1970 census-based measurements. In addition, the Census ’96 figures are close to those in the most recent available official birth-registrations for male and female children reported from hospitals and clinics for white, coloured and Indian children. (Unfortunately, under apartheid, the birth registrations for Africans were not fully recorded, and in 1993 the Department of Home Affairs decided that it would cease to record population group at birth. As a result, such comparisons are no longer available.)

It may be that both the 1970 census and Census ’96, even after adjustment for undercount, under-enumerated the number of males in certain age categories, especially among Africans in difficult-to-reach contexts such as hostels and informal settlements. But it may also be that the 1970 and the 1996 census measurements are largely correct, and the demographic models used in the intervening years hypothesized too-large proportions of males surviving in the African population group, at birth and subsequently. Further research on this newly-uncovered puzzle will be undertaken by Stats SA analysts in conjunction with their counterparts at universities and parastatals.

Table 2 below shows the respective population estimates broken down by province. The right-most columns show that, according to the final estimates, the shares of the total population have increased for three provinces, compared to the preliminary estimates. This involves mainly Northern Province (1,3%), but also KwaZulu-Natal (0,4%) and North West (0,3%). By contrast, the share has decreased mainly in Western Cape (1,1%) and Gauteng (0,8%). The other provinces’ share of total population is largely unchanged in the final estimates.

* All numbers given in this report are rounded to whole numbers. The totals may therefore differ slightly.

**    The percentages are rounded to the first decimal place, therefore they may not always add up to exactly 100.

The breakdown of the population into urban and non-urban place of enumeration, according to the preliminary and final estimates of Census ’96, is shown in Table 3, again compared to the 1991 demographic model as projected forward to 1996.

The table shows in the right-most column that 53,7% of the population was living in urban areas in October 1996, as against the lower 1991 demographic projection of 48,3%, and the higher preliminary estimate of 55,4%.

The preliminary report surmised that a greater rate of urbanisation had occurred from labour-supplying provinces to labour-absorbing provinces than was expected from the projections of the 1991 demographic model. This still appears to be so, when one considers the two tables in conjunction, but less extensively than was surmised.

A higher rate of urbanisation goes hand-in-hand with declining fertility rates, and the fewer people found in the country in 1996 than the 1991 demographic model would have suggested may, in part, be due to both these processes.

* All numbers given in this report are rounded to whole numbers. The totals may therefore differ slightly.

**   The percentages are rounded to the first decimal place, therefore they may not always add up to exactly 100.

The breakdown of the population into the four main population groups according to the final estimates of Census ’96 is shown in Table 4, compared to the 1991 demographic model as projected forward to 1996.

* All numbers given in this report are rounded to whole numbers. The totals may therefore differ slightly.

**    The percentages are rounded to the first decimal place, therefore they may not always add up to exactly 100

***  This category includes approximately 8 000 people, mainly in Northern Cape, classifying themselves as Griquas. It is excluded from the calculation of percentages.

The right-hand column shows that in October 1996 a larger percentage of the population was African, and a smaller percentage white, compared to the 1991 demographic projections. This may be due in part to the effect of increased, predominantly white, emigration (Stats SA finds that figures from foreign governments indicate that two to three times more people emigrate than indicated on the control forms gathered by South Africa’s Department of Home Affairs). It may also possibly be due to lower fertility and higher mortality amongst whites than modelled in 1991. But under-enumeration amongst the white population may have been higher than estimated even after the adjustments by the PES. In Census ’96, whites may have been difficult to reach because of suspicion of the process and the difficulty in finding sufficient white enumerators.

In addition, in Census ’96, the method of classification into population groups had changed compared to earlier years. In common with other countries, classification depended upon self-perception, rather than an apartheid-based legal racial classification. It is possible that some people who were previously classified as white either did not specify their population group or else classified themselves as African. This may also partly account for the lower-than- expected tally of whites.

According to findings from Census ’96, 958 000 people of a total of 39,8 million were not born in South Africa (this total excludes those found in institutions on census night and those who did not specify where they were born). Of the total population, 455 000 declared themselves as not holding South African citizenship. This may be an under-statement, since there is likely to be a degree of ‘self-naturalisation’, whereby respondents represent themselves as South African, particularly in the case of illegal immigrants.

The demarcation phase of Census ’96 assumed, in principle, that the entire country would be covered and divided into EAs. But in some areas this was not possible due to time constraints, and vacant land between settlements and villages in previously-undemarcated areas was sometimes left uncovered. Subsequently, ‘Project Eagle’ – which Stats SA undertook in collaboration with the Department of Land Affairs – covered the entire country, including vacant land. This project, undertaken for the Independent Electoral Commission during 1997 and 1998, entailed the capture of information onto a geographical information system (GIS) for use in planning the 1999 national general election.

There were more EAs in the GIS of Project Eagle (about 94 000) than in the census (about 86 000). Stats SA investigated this, and found 7 896 EAs without census information linked to them. Of these 7 896 EAs, a total of 3 496 were classified as EAs with no population or ‘zero EAs’.

Stats SA then tested the situation regarding the remaining 4 427 EAs by means of a sample of 167 EAs. Fieldwork indicated that 30% of these EAs were empty at the time of Census ’96, and were still empty at the time of checking. Another 5% were empty at the time of the census, but were populated at the time of checking. The people who had moved in after the census indicated that they had been counted elsewhere. In another 57%, the vast majority of those living in them indicated that they were counted in Census ’96. They were therefore not reflected as living in the correct EA, even though they were counted. A further 6% of the EAs in the sample could not be identified on the ground, even though they existed on the GIS. They require more investigation. Two percent of EAs in the sample were commercial farms in North West, and the fieldworkers were unable to gain access to them.

From this exercise it seems that those areas which were not demarcated in the census, according to the GIS, were either empty EAs, or were included in the census as part of other EAs. It is unlikely that any populated EAs were entirely missed during Census ’96. People or households that were missed were taken into account through the PES.

In this way, the post-census completion of the GIS has provided a useful accounting check on the specified adequacy of the coverage achieved in the count and adjusted by the PES.