Symposium 2001/54

24 September 2001


                                                                                                           English only


Symposium on Global Review of 2000 Round of

Population and Housing Censuses:

Mid-Decade Assessment and Future Prospects

Statistics Division

Department of Economic and Social Affairs

United Nations Secretariat

New York, 7-10 August 2001












The Case of South Africa*

Pali Lehohla **




A. Introduction. 1

B. Approval process for a census in south africa. 1

1.    Stakeholder consultations. 1

2. User consultations. 1

C. Strategic choices among data-collection methods for South Africa. 2

1.    Is a sample census possible?. 2

2. If a “sample census” is not possible, when should a full census next occur?. 2

3. Option of a six-month postponement, to March 2002. 3

4. The benefit of a 2002 census for SADC collaboration. 3

5. The verdict 3

D. Adapting new technologies to census operations. 3

1. Data processing. 3

2. GIS as a tool to censuses processes. 4

E. Maintaining census-related activities during the intercensal years. 6

F. Identifying and resolving problems. 7

1. Results. 7

2. Measures taken. 8

G. Post-enumeration surveys: are they worth it or not?. 9

1. Lessons from the 1996 census. 9

2. Conclusion from the pilot 9

A. Introduction

1.                  South Africa's first all-inclusive census took place in 1996.  In continuing the practice of conducting a census of the population every five years, as well as in conforming to the Statistics Act, a population census is due in South Africa in October 2001. Of all the countries of the world, 21 countries conduct population censuses at intervals of five years. In the last two decades in particular, seven countries with a population of more than a million people conducted a census every five years.  These are Australia, Canada, Finland, Islamic Republic of Iran, New Zealand, South Africa and Turkey. The rest are by and large islands in Oceania and each has a population size of far less than a million inhabitants.  Let us now look at the continental spread of countries that carry out censuses every five years.


2.                  In Africa, only South Africa has a five-yearly cycle of taking a census.  In North America, Canada is the only country that undertakes a census every five years. In South America, the Falkland Islands are known to have undertaken a census every five years.  Among the Asian countries, three have undertaken a five-yearly census.  They are Islamic Republic of Iran, Maldives and Turkey.  In Europe the following countries have had a five-yearly cycle: Channel Islands, Finland, Isle of Man and Sweden (although the Swedish practice was short lived). Oceania has the largest number of countries that undertake a census at five-yearly intervals—11 in number. They include Australia and New Zealand.[1]


3.                  The census in South Africa is undertaken within the context of harmonization of the statistics of the Southern African Development Community (SADC).  In this connection, the 14 member states have initiated an ambitious programme of harmonizing statistics in an attempt to influence and achieve measurable economic integration.  The census was among the first programmes aimed at harmonization. Milestone workshops and agreements, including one on the census questions, have been concluded among the states.  Eight of the SADC member states will have conducted or will be conducting a census of the population in the years 2000 to 2002.  The community nominated South Africa as the focal point for this activity.  The United Nations Population Fund (UNFPA), United Nations Statistics Division (UNSD), United States Agency for International Development (USAID) and the Swedish International Development Authority (SIDA) contributed generously to see through the implementation of this project.


B. Approval process for a census in south africa

4.                  South Africa, by act of Parliament, conducts a census of the population every five years.  In preparation for this, Statistics South Africa, (StatsSA), prepared a cost estimate of R623 million for this purpose, nominally an increase of  R190 million over the amount spent on the 1996 census. This budget was presented to State Expenditure for funding. However, State Expenditure allocated only R240 million for the activity then.  The amount of allocation was seen to be too little to execute a census, even by the 1996 standards.


1.      Stakeholder consultations


5.                  StatsSA subsequently presented the budget as approved to the Statistics Council, a statutory body that advises the minister responsible for statistics and the head of Statistics South Africa in its meeting of 22 April 1999. Council advised StatsSA to have a wider consultation with all stakeholders for options. The first consultation was with a reference group that aimed at giving shape to a technical discussion document (TDD).  As a follow-up to these consultations, the TDD was tabled to a larger group of stakeholders for discussions and recommendations in July.  Evolving out of the discussions and recommendations, the Statistics Council and StatsSA formulated a Cabinet Memo that sought government’s decision on the nature of a census in 2001.  The undertaking was subsequently approved and funds allocated in October 1999, giving StatsSA two years to plan and implement the census.  In terms of time, this marked an improvement of a year on the planning process compared to that of Census ‘96.


6.                  Justification for the next census is based on the strong demand for data at the local-area level. The public and the private sector, as well as government, require small-area statistics. This means that the demand for statistics goes well below the geography of provinces, district councils and local government entities.  In many instances a village or suburb level is what users require.   This is needed for purposes of targeting the clusters of population that are of interest.


2. User consultations[2]


7.                  A “road show” of user consultations was mounted from February 2000 to July 2000.  In this period there were about 16 workshops conducted with different stakeholders.  Each of the nine provinces hosted a provincial workshop to contribute to the content and possible framing of the questionnaires.  Below is an outline of the memo that solicited participation.  The memo also had a detailed list of questions and explained the intended purpose as well as the results as observed from the 1996 census.


8.                  This memorandum is intended for providers and users of census data and other interested parties in order to start a discussion on the contents of the census questionnaire. Although the census questionnaire used in 1996 worked well, there is still a need for revision of contents and formulations. The goal is to create a questionnaire that:


·        Decreases the time for data collection (shorter interviews);

·        Improves the quality of the data (more accurate answers, fewer unknowns);

·        Decreases the data processing time and costs (more pre-coding); and

·        Decreases the total expenditure of the census. The budget is only about R16 per capita, which is not much by international comparison. The enumeration in 1996 took three weeks. This time the goal is to complete the enumeration in one week to 10 days without sacrificing quality.


C. Strategic choices among data-collection methods for South Africa[3]

9.                  Financial constraints led StatsSA to have a dialogue with experts and other users on what option would be best for South Africa regarding the census. There were two options open to South Africa, one for a sample survey and the other for a census, but later rather than sooner. Below is an outline of the pros and cons and a final verdict. The assessment in the consultative workshop proceeded in several stages.


1.      Is a sample census possible?


10.              Firstly, it was considered whether a reduced or "sample" census was feasible and cost-effective and whether it would produce the desired information. If it met these three requirements, it would be the obvious way to proceed. However, it clearly emerged from the expert inputs, and somewhat surprisingly to the other participants, that the answer to all three questions was "no".


11.              To achieve a precise and unbiased sample requires that each enumerator area (EA) in the sample, a "patch" of 75-150 households, be "listed" in advance—i.e., the separate households have to be identified and noted. (This would have to be done in some 15,000 enumerator areas for a likely-sized sample.) This listing ensures that an appropriate sample of households can be drawn for the enumerator to visit, and that the enumerator can later be checked upon, to ensure that he or she does not bias the sample by skipping the difficult-to-reach households in favour of the easy ones. Orderly listing is easy to envisage in a formal suburb with addresses, but much more difficult in a rural area or an informal settlement.


12.              The user consultations revealed that these rather intricate processes are actually more difficult to conduct and control than a full census enumeration. In a full enumeration, one requires only the outline of the enumerator area—now available from the current Geographic Information System (GIS), although requiring improvement—and then instructs the enumerator to attempt to reach every household, and keep a record as he or she proceeds. As a consequence, the population estimates emerging from the sampling process would be more controversial. This is an undesirable feature when allocations of parliamentary seats and tax revenues depend upon them.


13.              Moreover, the usual means of cost-saving in a sample survey—that one has to visit only the sampled enumerator areas—falls away when one is seeking to achieve population estimates. Before the sample can be multiplied up to yield figures for the full population, all the “patches” that are not in the sample would also have to be visited and the number of households identified and counted. This would largely dissipate the notion of cost saving.


14.              This in turn affects the kind of data that are yielded. To update the GIS one wants a widely "dispersed" sample—i.e., one that visits as many “patches” as possible—but this minimizes the savings. Conversely, the savings improve as one visits fewer patches, but then one cannot use the resulting data to satisfactorily upgrade the GIS down to enumerator-area level.


15.              Worse, more than 700 of the present 843 local authorities have fewer than 4,000 households—equivalent to maybe 40 enumerator areas. If one sampled only half a dozen of them in each of these local authorities (which is all that could be afforded with a R240m allocation), it would clearly be very risky to take these half dozen as representing the variation across a whole local authority.


2. If a “sample census” is not possible, when should a full census next occur?


16.              With a “sample census” out of contention, and if only R240m were to be available for the three years, the next question was: Should a full census rather be undertaken later, when more funds might be available?


17.              The users were adamant that they did not want as much as a 10-year interval between the 1996 census and the next one. Some plausible arguments were advanced, however, in favour of a seven-year gap (that is, the next census to be in 2003 with data becoming available in 2005).  This argument notably was supported by the notion that many population features such as fertility or the education profile only change gradually, and that there would be more time for careful preparation.


18.              In favour of retaining the five-year interval (i.e., the next census in 2001), it was argued that yet other, very important population features—notably, the geographical distribution of AIDS-related deaths, and internal migration—were changing rather fast. Moreover, the Reserve Bank argued that, given the importance of population estimates in the five-yearly revision of the GDP (gross domestic product) in national accounts, a 10-year gap would be too long and a seven-year gap would be inconvenient.


3. Option of a six-month postponement, to March 2002


19.              Budget considerations and smoothing out the expenditure over two years appeared a compelling argument for postponement by six months. The six-month postponement appears to be the best option: a full census providing small-area data and GIS updates, rather than a methodologically risky sample census; with slightly more time available for better preparation; but taken soon enough to yield the fresh information demanded by user departments regarding fast-moving demographic changes.


4. The benefit of a 2002 census for SADC collaboration


20.              In addition, a census in 2002 would have the advantage of still including South Africa in the 2001/2 round of population censuses being conducted by more than half of the countries in SADC. There is a collaborative programme being led by StatsSA as part of its African Renaissance thrust, aiming to harmonize the core questions, especially those dealing with cross-border migration. Since so much migration affects South Africa, it is highly desirable that we conduct our census near enough in time to the others in SADC.


5. The verdict


21.              The last recommendation of a postponement by six months was thrown out in favour of keeping to a five-year interval as requested by law.  Thus, South Africa is poised to mount a census in October 2001, exactly five years after the previous one.


D. Adapting new technologies to census operations

22.              StatsSA has invested extensively in new technology, in particular, the GIS for managing and directing cartography and downstream activities of the census, including mapping for dissemination.  It has further adapted scanning for data processing.  These new technologies imply and demand new skills and different management tool kits.


23.              Adopting new technologies brings about several challenges.  One of them is the capacity of management.  The other is the potential threat of failure caused by less optimal synchronization between technology and those who manage it.  Managing an outsourced environment can be a daunting task, as we will illustrate.  The key problem lies with the management capacity of statistics offices to engage an outsourced environment effectively.


1. Data processing


24.              The management of a census is that of containing risk, largely because of its time- sensitive nature and subsequently of wasted capital if not undertaken at the scheduled date. 


a. Risk management


25.              Bringing on board new technologies can increase the risk profile of a census, particularly in the area of data processing.  We will explore some of the near horror stories in some countries in the current round of census taking. 


26.              Although Botswana will not be using scanning, they had to call off a tender on data processing and appointed a new vendor instead.  Those who handle scanning will attest to the fact that with it come issues of paper quality, print precision and shading contrast as very important considerations.  The capacity and competency of the printers to deliver on time is also a critical issue to consider, as this can constitute the single most important variable to undermine the success of a census. The outsourced companies are unfortunately not often exposed to these important facts and in their proposals may not be sufficiently open about their limitations lest they do not get the tender.  They are at times keen to get the business instead of focusing on the concerns of the statistics offices.


27.              Paper and pencil quality risk. Countries that have used scanning have horror stories to tell about their experiences either in the testing or during the actual undertaking of the main census itself.  Kenya’s data capture was delayed because of conditions associated with the use of scanning with some devastating consequences at the individual level. In the United Kingdom, the pilot census revealed that ink contrast affected readability, while the United States pilot showed that contracting to a multiplicity of producers compromised quality of print and therefore limited the ability of the scanners to read characters proficiently.


28.              Questionnaire delivery risk. In Zambia, for instance, the census was postponed by two months, from August 2000 to October of the same year, because the questionnaires for scanning were not delivered in time.  They were printed in the United Kingdom.


29.              South Africa has its horror stories too. Three months before the census, in June 2001, South Africa had to contract with a company in the United States to print half of the consignment of the questionnaires because the local producer would not print an adequate number to meet the required lead times for distribution. To date the South African component has delivered only 4 million questionnaires, since they started eight weeks ago, of the 14 million they were supposed to deliver, and they are not in a position to deliver the balance in the next two weeks.  The United States component has completed 7 of the 10 million, with the remainder coming by 14 August in time for distribution.


30.              Paper flow and identification risks. The testing phase of the scanning in April already revealed that there are areas that require serious attention.  The quality of printing was inadequate; the bar codes on the questionnaire at times were not machine-readable and the numbers had to be transcribed by hand, subsequently causing serious delays in processing.  StatsSA insisted that each page of the questionnaire have a bar code for purposes of identification and ensuring that sheets in a questionnaire are uniquely associated.  The imposition caused the printer to punch windows on the questionnaire to meet this requirement.  While this worked somewhat well in the field, the trailing threads of paper from the windows caused the scanner to jam often.   The processing of the questionnaires from the pilot is behind schedule by about four weeks when it was allocated 12 weeks initially.  


31.              Knock-on effects. With the printing in the United States, it has been necessary to fly some officials who were responsible for scanning to the US to continuously check and confirm quality.  The strategy has worked well but has caused further delays in the processing of the pilot census.


32.              Costs. The cost of paper for scanning can be prohibitively expensive and a comparison for South Africa suggests that the cost will be upwards of twice the cost of ordinary paper.  When you add the risk of speedy delivery and the associated freight charges, Zambia and South Africa should be licking their wounds.


b. Conclusion


33.              The scanning technology can be very efficient when handled with the necessary care—for instance, in the case of Turkey and later Ghana; indeed, the United States and the United Kingdom are success stories in scanning.  However, the possible problems that run counter to the critical path for census execution can be so many as to threaten the entire project. 


2. GIS as a tool to censuses processes


a. Background


34.              Census 1996 captured EA boundaries digitally, retaining verbal boundary descriptions on 02 forms. The EA boundaries were digital, but referred to a paper-based background. When the results were reconciled for Census 1996, it was found that there were a number of severe shortcomings in the demarcation. These included :

·        Inaccurate capturing of the boundary locations;

·        Unacceptable shapes and topology for the EAs;

·        Vague boundary descriptions;

·        No addresses or problems relating them to a specific position; and

·        Little or no relation of boundary descriptions to physical features on the ground.


35.              Demarcation for Census 2001 was an attempt to rectify these shortcomings; it meant that this would be a different ball game as compared with previous censuses. Several resources were considered to be of crucial importance in a solution of the demarcation problem. The resources were of two kinds:


·        Improved technology, not only better, but also more economical; and

·        The availability of data captured by other organizations.


In view of this, the census-mapping process depended on some fundamental vector data sets.


b. The EA backdrop


36.              Consideration was given to the acquisition of the following backdrop:


·      From the Department of Land Affairs: aerial photos, orthophotos and place name details;

·      From the Independent Electoral Commission (IEC): information on population numbers as provided by municipalities; digital information from the corporate database on EAs which were split;

·      From the metros and local authorities: aerial photography; and

·      From private companies and parastatals: the most recent mapping and aerial photography, mapped villages, satellite imagery, street names, etc.


c. Unresolved problems


37.              We argued for the latest technologies and the benefits they bring about, including cost savings and possibly planning to do away with the old way of doing things, in particular, foot-soldier listing and verbal descriptions of boundaries.  It was to our horror that we noted that listing and verbal descriptions of boundaries remain very important despite the emergent technologies. While the problem of identifying boundaries was to some extent associated with poor enumerator training on mapping, it was also clear that verbal descriptions, even in the presence of clear maps, enhance the usability of the maps by the enumerator. The pilot highlighted critical areas that we are desperately addressing with the people we will earmark as field supervisors. These relate to the following:


·        The clarity and legibility of the 09 map has been reported as problematic in some areas;

·        In some cases the place names are not correct or complete; and

·        Structures are not being correctly identified.


38.              We believe that when these issues have been well contained and controlled for in the next three weeks, we will be on the road to a successful census.  


d. Benefits of the approach


39.              In general, the technology itself has worked. The capturing of EA boundaries is almost complete. For each EA we have captured extensive EA metadata, such as types, entities, institutions, contact details and addresses. This information is put into a detailed database structure.


40.              Despite limited time frames to demarcate EAs, the technology was used to delimit Supervisor Units, (a Supervisor Unit is a grouping of five EAs within the same area type; a GIS utility combines adjacent EAs of the same area type). An exception report is generated, and the operator resolves all exceptions (similar to the IEC’s Voting Districts).


41.              EAs were re-numbered spatially. A GIS utility re-numbered adjacent EAs within a municipality.  These will help with geographic sampling for surveys where the assumption of spatial contiguity in numbering will now hold—unlike in the past where this assumption was thought to be true and indeed was false.


e. Risks and associated costs


42.              The process of arriving at the 2001 enumerator areas was broken down into smaller chunks. A GIS system had to be acquired, as the organization did not have the appropriate tools to perform the task that lay ahead. Hardware, which included high-end servers, personal computers (PCs), printers and networks, for both the head office and provincial offices, was acquired and installed. Due to the large volume of image data the GIS server is an incredible 1.8 TB system. Software for the appropriate GIS technology was benchmarked and acquired. Software included database software, GIS software and image-processing software. The contracts included software customization, database development, programming and training and four full-time GIS resources on site for two years for capacity-building of the StatsSA team. The system was up and running by the end of May 2000. The spatial database was rapidly being updated with data as received from various data suppliers. The GIS hardware and software cost in the region of R30 million. As the project grew and as the data grew, and as deadlines were threatened, the GIS system had to be expanded, resulting in more expenditure.


43.              The capturing of enumerator areas cost in the region of R9 million. The process followed up until the end of April 2001 was to make use of the local knowledge of provincial office staff to take the final decision on the EA boundaries. However, by May 2001, the time lines had slipped seriously (the sending of data and maps to provincial offices and the lack of skilled personnel at provincial offices to manage and demarcate EAs contributed most to slippages). This process was basically stopped and intervention was taken to allow the contractor to demarcate and capture directly on-screen. Two additional demarcation-capture sites were set up at the head office. One of the teams was made up of a “crack” team consisting of the best demarcators StatsSA had from provincial offices and the head office. The demarcation and capturing process was done by mid-July 2001. The intervention had cost the GIS project R5 million (from the R9 million mentioned above).


f. Conclusion


44.              Having used the up-to-date backdrop, technology and process during the 2001 demarcation phase, it is clear that census taking, especially census mapping, will never be the same. A review of the 2001 census-mapping process is currently under way, and it will provide invaluable inputs for the conduct of the next census in 2006.


45.              Preliminary indications are that most of the shortcomings being identified at this time could be attributed to the increased ability to compare the 09 maps with the terrain. Many of these shortcomings would have gone undetected in the past.


E. Maintaining census-related activities during the intercensal years[4]

46.              Statistics South Africa runs, by act of Parliament, a census every five years.   This unique position imposes substantial resource utilization at intervals that are very short; however, as currently constituted StatsSA is not in a position to sustain a five-yearly census cycle.  To execute this parliamentary mandate StatsSA needs a core structure to manage census activities.  This part of the paper looks in detail at the mandates and expected deliverables, as well as the associated staffing requirements.


47.              The activities for the census include planning, implementation and dissemination of the census results.  It is also required that such a structure manage its activities in accordance with the financial and labour laws of the land.  Therefore, there is a need to have a corporate services structure that ensures that there is compliance on deployment of financial and human resources. Although a census may be viewed as a project and historically has deployed resources for its execution only at the time or as near as possible to the time of the event, experience demonstrates that there are good reasons for retaining a core team that maintains best practice and continuity. It has been demonstrated and usually recommended that a census be planned optimally three years prior to the undertaking.  This has been the practice in most of the Third World countries and often the lead times could even be as short as 8 to 12 months. In the more developed parts of the world a census is planned in cycles of seven years.  For instance, Canada, which is conducting a census in 2001, had the strategic meeting for Census 2006 in 1998. The submission is a modest attempt to seek comment on the minimum functions that in our view would be continuous and thus require a set of skills that would guarantee implementation.


48.              This part of the document examines the mandates and deliverables of a permanent structure for the census.   It further looks into a set of tools and skills required for the performance of the tasks at hand and finally suggests a structure and speculates on the levels as well as numbers of people that would be needed. It builds on the current experiences of the census programme team.  It should be seen as a contribution to the current discussions on the mandate and new structure for StatsSA.


49.              The structure proposes that there be six components, with census operations being responsible for 67 per cent of the personnel.  The components are listed as follows:


Area of activity                                                 Staffing requirement


1.0       Project management                                                    6

2.0       Research, testing, quality and analysis                5

3.0       User consultations and content                          3

4.0       Census operations                                                      63

4.1                   Geography       31

4.2                   Collection         15

4.3                   Processing        13

4.4                   Dissemination   4                                             

5.0       Publicity                                                                       2

6.0       Corporate services                                                     14

Total                                                                                     93


50.              The structure is top-heavy, with 12 per cent consisting of directors, 64 per cent being assistant- and deputy-director-level appointments, another 11 per cent being statisticians and the remaining 13 per cent consisting of administrative officers. This composition of the structure reflects that a lot of the work will be done through short-term contracts, subcontractors and outsourcing.  StatsSA staff will be primarily informed managers of the process. The operations will largely be driven internally, and this is reflected by the more than 67 per cent of the people being in operations.


F. Identifying and resolving problems[5]

51.              Pre-testing of census content, methods, procedures and processes is very useful in providing a basis for decisions that should be made during the advance planning of the census. For majority of tests, it is usually important that as far as possible, the sample selected is representative of the whole country. The areas selected for Census 2001—although in many respects based on purposive methods—met this criterion. Another important point to note is that because of the time limit, the pilot could provide answers only on procedures and processes in enumeration with very little on actual content and data-processing procedures. The results, however, may not be generalized to the population because the sample as drawn was not probabilistic. However, the timing of the pilot census failed the test, as it was supposed to have occurred in October 2000 to coincide with October 2001, when the actual census was to be conducted. The pilot census was conducted in March 2001. Following the pilot, sets of evaluations based on experiences were instituted. They are as follows:


1. Results


52.              The results of the above exercises are summarized under areas that contributed highly to the poor performance of the fieldwork. They are:


·        Lack of organization and supervisory capability: Procedures not adhered to, flow of information uncoordinated, lack of professional accountability.

·        Size of the EAs above the norm (too many dwellings) and enumerators not being able to identify boundaries. Uncompleted EAs.

·        High number of refusals: Lack of publicity and poor calibre of enumerators.

·        High numbers of non-contacts: Too many empty holiday homes and to a lesser extent poor management of callbacks.

·        Problems in administering the questionnaires: Many questions not attended to. 

·        Poor calibre of field staff associated with unclear recruitment policy.

·        Poor training of field staff resulting in questionable output across the board.

·        Listing procedures not uniform: Total number of dwellings/households different  between PES and census enumeration.

·        Lack of proper quality assurance procedures during enumeration; poorly filled-in questionnaires.


2. Measures taken


53.              The following steps have been taken to address the above problems in enumeration:


§         One experienced officer from household surveys has been seconded to census as chief operations officer to help in planning and implementation of activities both at the head office and in the field.

§         Four experienced technical staff from PES, census mapping and household surveys have been seconded to enumeration. Three of them will look at the listing procedures in detail, and one will concentrate on the enumeration procedures and methodologies and development of manuals.

§         There is currently a move to reorganize and strengthen the provincial offices by having project management experts attached to their operation base.

§         The whole recruitment strategy is being scrutinized with the view of employing experienced regional managers who in turn are going to work in conjunction with the StatsSA regional surveys managers as advisers on data-collection techniques. The recruitment will also be decentralized for close supervision and implementation.

§         The listing strategy has been revised and listing is going to start in June rather than two weeks before enumeration. This should provide indicators on enumerators’ workload well in advance.

§         During listing, verbal boundary descriptions of EAs will be documented.  This will assist the enumerators in locating the confines of their EAs. Furthermore, census-mapping staff are going to be deployed in both the initial stages of listing and also during actual enumeration so as to maintain accuracy.

§         Currently, the publicity personnel are busy launching the integrated publicity programme in all the provinces. The activities are going to run through to enumeration and hopefully, this will help in raising awareness and in turn improve on the enumeration acceptability levels among the difficult groups.

§         To maintain a high degree of professionalism and maintain quality, a decision has been taken to have predominantly teachers as supervisors.

§         In the current recruitment strategy, the requirements in terms of qualifications, experience and locational factors for the fieldwork personnel are being correctly put into place. The monitoring of the process will be from head office, provincial and regional offices. In the pilot census, procedures were loosely defined and thus allowed for mediocrity.

§         The enumerators’ and supervisors’ manuals currently being developed have strictly spelled out the procedures to follow in the management of callbacks to avoid laxity in the subsequent follow-ups (no one was sure in the pilot that there were any follow-ups). In other words, the refusals will be a joint responsibility of the supervisor and enumerator closely monitored by the fieldworker coordinator.

§         The training strategy has been drastically reviewed to make sure that 80 per cent  of the enumerators’ training is on the census questionnaire unlike in the pilot, where the questionnaire was given only 10 per cent of the training time.

§         On the other hand, there has been an introduction of training videotapes geared towards efficient and uniform messages across the board.

§         The whole recruitment strategy has been revised to accommodate the recruitment of field staff at the regional level. This will assure that the calibre of the field staff is of the required standard and maintain the policy of having field staff from their area of operation. An allowance has been made to have the whole process monitored by a wider team from the head office and provincial office to eliminate possibilities of nepotism and so forth.

§         The time to be taken on the recruitment has been increased considerably also to allow for intense scrutiny of the target staff.


54.              The training strategy has been reviewed and the following procedures have been put into place:

§         The national trainers are going through intensive in-house training to bring them up to par as far as the subject-matter areas are concerned. The programme started immediately in April and will continue through June.

§         The national trainers are going to train together with the subject-matter specialists, at least at the first level of training.

§         The national trainers have been integrated with the enumeration team so that they will be able to understand the whole planning process.

§         A training videotape has been introduced as a training aid or guide for both the trainers and the trainees.

§         Training at the regional level will be carried out by the trainers and the regional survey managers who have vast experience in data-collection techniques.

§         A strong monitoring and evaluation team from the head office (with both internal and external monitors) has been proposed.The team will recommend immediate action if the procedures are found to be wanting.

§         Two experienced officers have been seconded to the listing project as a way of building the project base and improving on performance.

§         All the listing procedures are currently being reviewed with the main aim of standardizing them not only from the enumeration point of view but also from the standpoint of the PES and household surveys. The unit of listing this time will be the same across the board.

§         Listing will be carried out for a longer period of time and by a completely experienced and independent team of listers drawn from across the country.

§         All the fieldwork procedures are being reviewed by a team of experienced personnel with the aim of improving the management of questionnaire content and overall fieldwork supervision. The role of supervisors and enumerators will be more closely monitored with definite measurable indicators, unlike the pilot.


G. Post-enumeration surveys: are they worth it or not?

55.              In South Africa prior to the census of 1996, there is evidence of adjusting census information at least for the censuses of 1980, 1985 and 1991.  The only well-documented adjustment process is that of the census of 1991.  In the case of the other censuses, one can infer that there were adjustments by comparing published information with the original records.  In the census of 1991 the adjustment was based on a demographic model and not a PES. 


56.              The first attempt at a PES was carried out in the census of 1996. This has been well documented and it includes limitations.  


57.              South Africa has a policy position of a “one-number” census. In this regard, the census count is adjusted by the PES information and the resulting adjusted population is then published for public use.  A 10 per cent sample of individual records is also made available to the public.  In the census of 1996, South Africa had an undercount of 10.4 per cent, which was duly adjusted by geographic level and by race, among other factors.


58.              The census figures are used for the allocation of funds to provinces and determination of the resource split among three spheres of government—namely, national, provincial and local governments (municipalities).


59.              In the census of 2001 in October, a PES is planned and will be mounted. An initial report on how well we fared in the census using the PES has drawn our attention to several key areas of focus, as discussed below.


1. Lessons from the 1996 census


60.              Several limitations of the 1996 PES that are listed in the population census report were addressed by the PES test.   These included the following:


§         Methodologies and procedures were tried out and revised throughout the process;

§         The questionnaire design did not make adequate provision for those households and individuals that have moved into a new EA after the census;

§         The exclusion of the full names of people on both the census and the PES questionnaires, to enhance confidentiality in politically tense areas, made the matching process difficult;

§         The matching process was especially difficult in those areas where there were no addresses; and

§         Unclear EA boundaries, resulting in the inability to find the matching questionnaires, were the main reason for the unresolved cases (22 per cent) that then required statistical imputation.


2. Conclusion from the pilot


61.              Reduction of unresolved cases is attributed to use of names and stickers.  The names might have had a larger contribution in this compared to the stickers since sticker retention was not that good in some provinces.  But the idea of stickers should not be underestimated since stickers help more in identifying and matching questionnaires while names help more in identifying and matching individuals in a questionnaire. What comes out very strongly is that people do not understand the concept of stickers, and therefore they need some explanation in the form of publicity. Fieldworkers themselves had such problems as forgetting the sticker altogether, handing more stickers than necessary to one household and so forth. Therefore, use of both is highly recommended, provided good publicity about stickers has been done and a well-thought-through approach to training fieldworkers has been applied.  In addition, numbers on the aerial photographs would definitely assist further in matching.


62.              In CHAID analysis and when using proportions, the two types of enumeration do not appear to be different from each other in terms of number of people included in the census. It is up to the management team to take a decision on this, but other factors will have to be considered—for example, what the decision means in terms of sample size, cost, accurate listings and so forth. However, in consultation with sampling statisticians from the US Census Bureau, they recommended against sampling within an EA.  According to them, subsampling would complicate the field operations and would make it impossible to identify households that were included in the census but not in the PES during the matching process. 



*       This document was reproduced without formal editing.

**     Statistician General, Statistics South Africa. The views expressed in the paper are those of the author and do not imply the expression of any opinion on the part of the United Nations Secretariat.

[1] Technical Discussion Document (TDD) on options for 2001: “Census or Sample Survey”.

[2] Excerpts from an internal document titled “Census Users Topics”.

[3] Summary of recommendations from the TDD workshop (July 1999)

[4] Proposed structure for the census department details.

[5] Pilot census summary report.