SA Census 2001: How much progress?

Lawrence Schlemmer and Britt Youens evaluate the latest census and find it wanting in some respects.

Summary - While Census 2001 is a major achievement, certain aspects of the results raise important questions. The estimated total population of 44,8 million seems broadly correct, and the data on inter-provincial migration are particularly useful. The figures show significant educational progress, a decrease in the back-yard shack system, and increases in the numbers of homes with electricity, toilets and phones. All of this is useful and plausible information. However, there are problems with the statistics relating to gender, age distribution and unemployment. The sex ratio of 91,7 males to 100 females is unusually low for a developing society and reflects a mysterious deficit of over 800 000 males between the ages of ten and 44. Is this because they were undercounted (perhaps because of the mobility of young men), or have males succumbed in greater numbers to diseases of poverty, as StatsSA suggests? This is a serious problem that requires investigation. If the deficit is real, it could lead to further erosion of family structures in poor communities and have a major impact on the economy. If the deficit is the result of an undercount, it needs to be corrected. The statistics also indicate an unexpectedly rapid decline in fertility, with 12 per cent fewer births than 10-15 years ago. Again, it is important for forecasting to know whether this is real or reflects an undercount. The census shows an alarming increase in unemployment from 34 per cent in 1996 to nearly 42 per cent. StatsSA questions this, citing its own 29,5 per cent estimate. However, StatsSA classifies a lot of sheer survival activity such as occasional car-washing as employment. Moreover, the census figure tallies with the estimate produced by a department of labour survey in 2000. StatsSA needs to accept the serious problem of underemployment that is so severe as to be tantamount to unemployment. The decline in the number of whites, from 4,43 million in 1996 to 4,29 million in 2001, is also problematic. Excluding emigration, the white population should have grown to about five million by 2001 and therefore there is a deficit of some 700 000 whites. This indicates either serious undercounting or emigration on a massive scale, and it poses a major problem for commercial agencies. Another puzzle is a rise in the numbers of whites with little or no primary education. Population growth is estimated at around 2 per cent per annum, which contradicts assumptions made about Aids mortality. This rate is crucial for forecasting and planning and therefore we need to ascertain whether the census figure is correct or requires adjustment. Other problems include an inexplicable increase in agricultural employment, and an increase in household income that suggests national GDP growth of 3,1 per cent for the period, which is far more optimistic than Reserve Bank estimates. However, census income statistics are suspect at the best of times. Thus the census is useful but should be treated with caution.

Census 2001 is a very large achievement. Very few people realise how difficult it is to conduct successful censuses in developing countries. Most of the results also add value to our understanding of our society, its economy, transformation and the effects of governance and state delivery. In contrast to the response to the initial results in 1996, this time we may not hear protests about missing workers from the labour movement and grunts of utter disbelief among demographers and economists. Hence we will also be spared the national embarrassment of a reluctant withdrawal of the first results and their replacement by re-weighted figures.

At the same time, however, aspects of the results and of the whole approach to the task of gathering national statistical information raise some rather big questions. The media have so far reported results without any evaluation of them aside from reporting on problems acknowledged by Census officials and the Statistics Council. There has been some questioning of whether or not the census operation is cost-effective, but once again arguments have been telegraphic.

At the time of writing the Statistics Council has yet to deliver its final verdict on the census. If it accepts the results, any qualifications notwithstanding, the results will become benchmarks and the final authority on a host of issues. But is this good enough, and why should we be concerned if it is not?

What the census tells us

The South African and provincial populations

Unlike the first results of Census 1996, the estimated total population of 44,8 million in 2001, as weighted to correct for the rather large undercount, seems broadly correct. With a little pushing and shoving it can be broadly reconciled with important comparative indicators like the 1999 voter registration and 2001 enrolments at educational institutions.

The provincial populations also make sense and the data on inter-provincial migration are particularly useful. They issue fair warning to Gauteng and the Western Cape that they are going to face formidable challenges in absorbing a flood of dirt-poor in-migrants in the years that lie ahead.

Socio-economic progress

The figures reveal significant and plausible educational progress, with the proportions of the population with grade 12 or higher levels of education having risen from 22,6 per cent in 1996 to 28,8 per cent in 2001, and an overall drop in the proportion of people with no education from 19,3 per cent in 1996 to 17,9 per cent in 2001.

Other indicators of social progress include the following:

  • A fall in the highly exploitative backyard shack system from 6,6 per cent of dwellings in 1996 to 4,8 per cent in 2001.
  • A drop in the proportion living in traditional dwellings — somewhat ambiguous progress because traditional dwellings have far better insulation than the thin-walled little box RDP houses that have been erected in rural areas.
  • The percentage of households with electricity for lighting has increased from 58 per cent in 1996 to 70 per cent in 2001, and the proportion using wood for cooking has dropped from 23 per cent to 21 per cent.
  • Houses with flush or chemical toilets have increased from 50,5 per cent to 51,9 per cent (but houses with no toilet at all have risen from 12,4 per cent to 13,6 per cent).
  • Although it is an expensive mixed blessing for poor households, the census documentation give the proportion of households with either landline or cellular telephones as having increased from 29 per cent to 42 per cent, with the largest increases in those provinces in which food security and sheer survival are much more serious issues than the payment of telephone accounts.

These are extracts from the plausible, largely unproblematic and highly useful information from the census. It is a pity, therefore, that other findings cast some doubts on the quality of the overall results.


The main problems include the gender statistics, some aspects of the age structure that indicate skewed results, estimates of unemployment as compared with the Stats SA Labour Force Survey findings, aspects of occupational and industrial structure, income shifts and what they seem to imply for GDP growth and the size of the white population.

Gender and age

One of the most generally accepted demographic realities is that as societies develop and life expectancy increases, more and more women outlive men. This means that the sex ratio — the ratio of males to females — falls below a parity figure, which is usually expressed as a sex ratio of 100. In developing societies, on the other hand, the rigours of multiple childbirths among poor women and the generally low life expectancies combine to produce sex ratios of around 100 or more men than women.

In the census results for 2001 the sex ratio is 91,7 males to 100 females — even lower than one finds in most developed societies. Part of this anomaly is the considerable longevity of our women but part of it is a serious and mysterious deficit of males from middle childhood up to the age of 40 to 45 years.

Inspecting the age-gender pyramid for 2001 shows that from age 10 up to the age of 44 there is a deficit of males relative to females of over 800 000. This is the age range in which demographers usually find parity or a slight surplus of males. The question is whether the census has significantly undercounted males and not been able to correct the undercount, or whether there exists some national pathology affecting male children and working age adults.

Chairperson of the Statistics Council, Dr Hilary Southall, has attributed the undercount to the mobility of young and working age men — thereby implying that they were undercounted in both the census and the post-enumeration survey that is used as a basis for correcting for any undercounts. Officials at Stats SA added a view that poverty causes greater health problems in male babies, boys and men, referring to “research studies”. There may indeed be some research indicating this, but we should firstly be aware of the danger of attributing causality in complex issues to only one variable like poverty. Unfortunately, this line of thinking reminds one of some of the other things, like Aids, that, at high levels, are blamed largely on poverty in South Africa. And then, secondly, we should ask if the answer lies substantially in poverty, why have the same patterns not been seen in all other poor countries?

The 800 000 or more men apparently missing from our statistics is not a problem to be taken lightly. One consequence is the erosion of the already disrupted family structure in poor communities and its consequent effects on children. It also poses some difficult questions on the effects of this loss to the South African working population. If there is a poverty-induced male health crisis it needs to be investigated in depth and not simply used as a convenient defence of census results. If however, the male deficit is due to a failure of coverage in the census, it needs to be officially acknowledged and corrected. So far the reactions to this issue have been as problematic as the pattern itself. We assume, however, that the Statistics Council will interrogate worrying patterns such as this one and if necessary recommend follow-up analysis.

Lastly, there is also the fact that the results, on the face of it, indicate an unexpectedly rapid decline in fertility, with the age structure showing at least 12 per cent fewer births in the five years before the census than ten to fifteen years ago. If it is a real decline this is important to know but if it is an uncorrected undercount it will prejudice all population forecasting. Here again this bias would need to be acknowledged so that demographers can compensate for it.


The unemployment rate, according to the census, increased from 34 per cent in 1996 to nearly 42 per cent in 2001. Alarmed by this, the reaction of Stats SA has been to question the validity of the census figure and refer readers to the rate of 29,5 per cent estimated by their 2001 Labour Market Survey.

We should not simply accept that the census figure is all that wrong. Firstly it is close to the estimate made in the very large Metsebetsi Labour Market Survey commissioned by the Department of Labour in 2000. Secondly it is higher than the Stats SA Labour Market Survey estimate because the latter classifies a great deal of sheer survival activity on the streets as “employment”, backed by ILO conventions of doubtful applicability to labour market situations such as ours. An unemployed person who manages to wash two or three cars a week feels himself or herself to be unemployed, with utter justification but is likely to be classified as employed in the Labour Market Surveys. The reality of people’s perceptions of their employment status is what is reflected in census 2001.

There is some serious undisclosed income among people who call themselves unemployed, but it is a minor problem compared with unemployed people who have to do odd jobs to eat. Rather than deflect attention away from the finding, Stats SA should accept the serious problem of under-employment that is so severe as to be tantamount to unemployment. It should point out that the more precise definition of unemployment adopted in the Labour Market Surveys excludes a massive amount of survival activity by people who are effectively unemployed.

Missing whites: high walls and Rottweilers or emigration?

Whites are shown to have declined in numbers from 4,43 million in 1996 to 4,29 million in 2001. It has become urban legend among the middle classes that many people in richer suburbs tended not to be counted in 1996 and were missed again in 2001. If this is nothing more than dinner party talk, however, then there must be emigration on the scale of an exodus. The number of whites in the un-weighted (and therefore under-enumerated) count in 1985 was 4,57 million. Taking this underestimate as a conservative baseline and assuming that it grew by a very modest half a percent per annum until around 1993, the total by 2001 would be some 5 million without emigration. Has emigration decimated the white population, cutting it by at least 700 000 people, or is there a serious undercount?

Southall of the Statistics Council has offered the explanation of high walls, dogs and a disinclination in the middle class white suburbs to interact with census enumerators on the other side of intercoms. According to figures quoted by Stats SA one has to add another factor, namely that only 600 000 forms were printed in Afrikaans and those were used mainly in the Western and Northern Cape. In other words Afrikaans-speaking whites had a “constitutional” reason to be un-cooperative.

Given this manifest evidence of a white undercount, why was it not corrected or at least acknowledged and formally raised as an issue as in the case of the unemployment figures? Given the scope of white disposable income it is a serious problem for commercial agencies. A bank that has to be nameless found after the 1996 census that it alone had more bank accounts in one upper-middle class enumerator area than the number of households that the census had counted, and this kind of problem may recur.

There is at least one other gremlin in the white data. The numbers of whites with incomplete primary education or no education whatsoever increased between 1996 and 2001 from some 76 000 to nearly 105 000. Whites are perceived as a category that is preserving its privileges, and it is therefore very surprising that an upward shift in this category should occur.

The population growth rate

The growth in the census populations between 1996 and 2001 is estimated at around 2 per cent per annum. This is much higher than the growth rates that demographers have been using for strategic forecasts and it also contradicts the assumptions made about Aids mortality. For dozens of different reasons it is crucial that the figure of 2 per cent growth be confirmed or adjusted. Some demographers suspect that one reason why it could be misleadingly is that the 1996 census was a more serious undercount than the post-enumeration survey conducted by Stats SA indicated. The country needs answers on this as a basis for a great deal of forecasting and planning.

Other problems in the data

For the purposes of brevity, a few other problems in the findings will simply be listed:

  • Employment in agriculture and forestry is reflected as having increased since 1996, whereas employment in manufacturing, mining and public and community service shows an expected decline. Given the pressures on farmers to rationalise their labour and a host of evidence confirming this, the suggestion that agricultural employment is increasing, even if one includes informal and family employment, boggles the mind.
  • Census 2001 shows a welcome increase in the proportions of people in higher level and white-collar occupations, as one would expect. Unfortunately it also shows a massive decline of nearly 44 per cent in professional occupations. Either there were classification problems or there is serious bias in the coverage – it could be related partly to a white undercount.
  • According to a member of the Statistics Council quoted in the media, the household income data in the 2001 census compared with 1996, suggests a GDP growth per household of 7 per cent after inflation, and a consequent national GDP growth of 3,1 per cent per annum in the period (The Star, 9/7/2003). Economists pointed out that these figures were massively more optimistic than the estimates made by the Reserve Bank and the department of finance. But census income data is hugely suspect at the best of times. After the 1996 census Dr Hirschowitz of Stats SA was on record as warning that expenditure data should where possible be used in place of census income data. One trusts that the Reserve Bank and the Treasury will exercise due caution before they alter their macro-economic framework.

For these and other reasons not mentioned, the 2001 census results might warrant a gentle warning: “useful, but treat with caution”!

Is our census strategy appropriate?

Statistics SA places considerable emphasis on its task of recording progress in service delivery and living conditions. This emphasis burdens the census with a large number of questions and tends to dominate the presentation of results. As a consequence there is sometimes a flavour of public relations and promotion of government in the whole exercise. The long census questionnaires create great scope for bias and breakdowns in data collection. On this the statistician general has admitted that 95 per cent of the completed returns require face-to-face interviews. Is this an appropriate approach in a national census?

There are easier, cheaper and more precise ways of collecting data on levels of service delivery — large national surveys like the discontinued October Household Surveys are a case in point. Additionally, very brief questionnaires concentrating on essential aspects of population, its characteristics and distribution would relieve the census itself of the burden of training and supervising such large numbers of enumerators and would allow teams to concentrate on the accuracy of questions of core importance to a census.

The enumerators, however, are an issue by themselves. It is the experience of many researchers that the last people that should be used as survey interviewers are the unemployed. During the counting period there were reports in the press of many or some of the 85 000 enumerators employed walking out and having to be replaced at short notice, and of protests about shoe allowances, payment procedures and the like. This cannot be an effective situation.

Unemployed and ad hoc recruits simply do not have the social skills and the authority to get responses in areas that may pose various difficulties. To stay with the middle class area example, we have spoken to enumerators in these areas who told us that they waited to speak to domestics rather than to deal with suspicious and privacy or crime-conscious householders.

The 2001 census had an estimated 17,6 per cent undercount, up from 10,5 per cent in the far from perfect 1996 census. This meant that a huge burden was placed on what is called the post enumeration survey to establish the undercount as a basis for weighting and correcting the raw census results.

There are two problems with post enumeration survey-based corrections. They are statistically defensible for a country as a whole and for large regions and provinces. But because post enumeration surveys cannot cover local areas in detail, over-arching weighting based on such surveys can seriously distort local populations that might have been well counted, or very badly counted, in an otherwise moderately undercounted wider region. It is virtually impossible to weight up a census with a 17-18 per cent undercount to produce valid local statistics.

The second problem is that post enumeration surveys done by census teams at the end of an exhausting census tend to replicate the biases in the census itself. If high walls and Rottweilers discourage enumerators they also discourage post enumerators.

Given the fact that surveys can establish the things that the government wants to know about its performance in the field of service delivery, a census has to be directed at getting information that is essential for planners, engineers, marketers, economists, demographers and the treasury. This means descriptive population and household counts that are as correct as possible down to local and even suburban levels. It implies further that the questionnaire must be short and uncontroversial, that civil servants seconded from other departments and local government must be deployed as enumerators, and that the post-enumeration survey must be undertaken by agencies independent of the census itself. Only then will we be able to claim with credibility that we have “counted everyone in”.