Census 2011 results ‘fit for use’

Independent Newspaper Limited

Statistician-General Pali Lehohla and President Jacob Zuma during the presentation of 2011 Census results held at the Presidential Guest house. Picture: Thobile Mathonsi.

Related Stories

• Experts address census worries
• Census jobless data ‘came in too high’
• Fertility rate absent in data
• Census debunks myth about white men
• Urbanisation to boost SA economy

The results of Census 2011 have been released and they are being shared, explained and critiqued by a variety of stakeholders. Generally, the results have been well received as they shed light for policy action and form the evidence for hard choices that have to be made both at the political, business and individual level.

They are the template of evidence from which all and sundry have to draw. The results increase the space for consensus on issue identification, description, analysis as well as rationale, for options and solutions. They provide the geographic specificity of a range of issues raised.

A census undertaking is logistically very complex and its complexity is not mitigated because it measures human endeavour, itself a very complex terrain. Statistical information is declaratory and the consistency of individual declaration across space and time combined and corroborating independent and improving evidence from administrative records such as births and deaths, confirm the reliability of this declaratory exercise.

As South Africa received and engaged with the results, two professors from UCT, who were also members of an expert evaluation team for the Statistics Council, also engaged and commented on the results.

On October 22 I responded to their concerns. So for the record I should share what my responses were to their concerns. They raised five concerns and this is how I dealt with each of those then, and this is how I deal with them now.

From the work that the brain trust of Statistics SA had done and from the continuous feedback since May that they, the experts of UCT, and the 10 other experts made on the census, I presented the results of the census to the Statistics Council seating of October 20 to 22 for consideration with a mandate for their release.

First, on the question of receiving data late: I should immediately concede that I delivered the data or versions thereof later than the stipulated date each time to all experts, but the key difference was that the other 10 experts of the council were in situ at the processing centre, a key success factor for delivery, while the two UCT professors were not, and this was for good reason as they have full-day jobs.

The UCT professors argued that the count of 51.7 million was higher than their modelled estimate of 50.4 million. My response was that the adjustment for the count is based on a statistical, well-established sampling procedure.

In this regard, both Stats SA brain trust and the Post Enumeration Survey (PES) experts have confirmed that the sampling method was sound and the estimation procedure followed the statistical procedures in sampling. I further advised the UCT professors that their estimate and the number to which the count has finally been adjusted to fell within the confidence limits.

On the basis of the sampling procedure followed and the estimation processes executed, the result is sound and there is no statistical reason for them to express doubt in the results.

The UCT professors are not happy with the provincial distribution of the population since it does not match the 2001 census base after taking into account births, deaths and migration. My response to the UCT professors is that only under conditions of a closed population can their idea of a balancing equation work, and there are no managed borders between provinces and their argument therefore has no basis.

A knock-down argument is that the census count was conducted in 103 000 units which are mutually exclusive.

This was done after listing units of approximately 150. Each unit was enumerated independently. The aggregation by province arose from independently enumerated units. It is true that there would be deficiencies in the process, but there is no evidence of bias in favour or otherwise for any one of the provinces.

The PES was an independently conducted survey which attested to a differential undercount by province, further providing evidence that the 103 000 units had no aggregation bias towards a province.

Given that the estimates, both of the UCT professors and those in the census as adjusted, fall within the confidence limits, then by what scientific procedure would the UCT professors arrive or confirm their concern, or by what studious investigation would they navigate their enquiry and what scientific standing would such an enquiry claim? How would its findings be applied?

The UCT professors were not happy with the average number of people in households. Regarding the number of people in households, I pointed out to the UCT professors that the definition of a household and aggregation of persons as enumerated into households, is what is to be taken into account. The use of a questionnaire as a proxy for a household is not sufficient and will necessarily yield a lower average household size. However, we also confirm that the average household size in South Africa has been declining. By correcting for number of questionnaires used per household, the average household size increases, but just a little bit. The household surveys Stats SA has conducted, yield an average household size similar to the one in the census following the definition of what a household is.

The UCT professors are not happy with the edits. I have published the imputation and deck outcomes. The decks are less than 5 percent and in terms of the procedure for data processing as defined in the UN manuals, the data has high integrity. The decks do not distort the original data and the tabulation attests to this. The most important variables for the decks are 2 percent for age and 1.8 percent for sex. The acid test is whether these percentages are statistically significant and the answer is, they are not. The levels of imputation or deck in the data are far lower than the upper limit of the standard, and therefore the report on edits show high integrity of the data.

The UCT professors argue that the age structure, particularly five to 19 show a dip that is demographically untoward. Dr Feeney did justice to this pattern by confirming that it exists. Five independent observations have confirmed the existence of the pattern. First the Census of 2001, registered births, registered deaths, Census 2011 and PES have all independently affirmed this pattern. It exists in other jurisdictions and therefore it is not unique to South Africa. At this stage it is not clear as to what causes it and we can only speculate, but this does not deny its existence. It is the task of scientific enquiry and research to unravel its source and I look forward to further work among the scientific community of South Africa and the world to examine the genesis of the dip.

The UCT professors are not happy about the release date of the results and they argue that they are rushed. While the call to express an opinion to publish or not is not mine, but is statutorily one for council, I will however state that what I have undertaken as the statistician-general is to address each of the concerns the UCT professors raised and having done so, I find no reason why I cannot in presenting the results to council, conclude that the results are ready to be released and fit for use.

* Pali Lehohla is South Africa’s statistician-general and head of Statistics SA.