Back to contents index

Chapter 2

Methodology used in Census ’96

In this chapter, an explanation is given of the overall methodology used in the 1996 population census and its various phases. This includes:

• Pre-enumeration.

• Enumeration.

• Processing of data.

• Development, dissemination and marketing of products and services.
  

Planning, funding, demarcation, publicity, questionnaire design, printing, recruitment and training for Census ’96 had to be conceived and undertaken in a period of 15 months, because users of data for planning future development throughout the country could not countenance its postponement beyond October 1996. Elsewhere, the run-up period for a national population census is typically three years.

In order to ensure that a wide variety of stakeholders and users participated in the census, an advisory committee was set up consisting of diverse interest groups. These included academics, representatives of civil society, other government departments, parastatal organisations, universities, non-governmental organisations and the business community. This committee met at approximately six-weekly intervals throughout the preparatory and enumeration phases of Census ’96.

The members gave detailed advice on all aspects of the census, including questionnaire design, demarcation, publicity and enumeration. In addition, they played an important role in monitoring fieldwork during both the pre-enumeration and the enumeration phases of the census.

Sub-committees were also formed to give advice on testing the questionnaire, and later to help plan census products.

Extensive assistance regarding strategy, demarcation, fieldwork, training and data processing was given by census specialists from the Australian Bureau of Statistics, Statistics Sweden and the United Nations Fund for Population Activities (UNFPA).

The processes of demarcation and listing are closely related. Demarcation divides the country into small geographic areas with clearly distinguishable boundaries, called enumerator areas (EAs). Listing, on the other hand, takes place within each demarcated EA. It involves making a list of all visiting points within that particular EA, and indicating where to find each visiting point.

A visiting point is defined as a distinctive site, stand, premise or property containing one or more dwellings. It is a clearly-distinguishable place that the enumerator is required to visit in order to administer or deliver a census questionnaire or questionnaires. Each dwelling at a visiting point can contain one or more households. For example, on a given property, there might be a house which is subdivided and occupied by two families, one of which employs a domestic worker living in a backyard room. In this case, there are three households in two dwellings at one visiting point.

In a formal urban area, visiting points are usually associated with a specific street address. In an informal settlement, however, very few visiting points have specific addresses, and numbers were painted onto the shacks during the pre-enumeration phase, where possible, to make them more easily identifiable for enumeration.

In a non-urban area, the visiting points were also unlikely to have specified addresses, and these had to be identified by a route description or householder’s name. In yet other areas, particularly remote non-urban areas, listings by household name were completed with the help of the local headman or tribal authority, or the name of the farm was used. In a country where perhaps half of all households do not have a specified physical address, the listing of visiting points was not easy to achieve.

The country was divided into approximately 86 000 EAs. Each EA consisted of about 100 to 250 households, the size of the EA being determined by whether its location was in a densely or sparsely populated area. The underlying principle was that all parts of an EA should be within comfortable reach of an enumerator, or person doing the counting.

The information on each demarcated and listed EA was brought together into a booklet called the census summary book. This booklet included maps, aerial photographs or descriptions of the EA and its boundaries, and the address or other means of identification of each visiting point and the households within it. In addition, space was left for completion during the enumeration phase regarding whether or not the household had been enumerated.

• If the household was not enumerated, the enumerator was required to give a reason why this happened. For example, there could be a refusal, a non-contact, an unoccupied dwelling, a vacant stand or a non-residential property such as a shop, church or school.

• If the household was enumerated, the enumerator was required to indicate the number of people in total, and the number of males and females in the household.

In 1991, large parts of the country were not demarcated into EAs, which made the task of demarcation especially difficult in 1996. In particular, the following were excluded from the 1991 demarcation process:

• The former ‘TBVC’ (Transkei-Bophuthatswana-Venda-Ciskei) states. These did indeed have separate censuses conducted in each ‘state’ in 1991, but these were based on different methodologies, and counts were never integrated into the total South African census.

• Rural areas of the erstwhile ‘self-governing territories’, most of which were counted in 1991 by means of a ‘sweep enumeration’ covering large areas of land.

• Squatter camps and informal settlements in urban or semi-urban areas, for which population estimates were derived from aerial photographs.

For Census ’96, on the other hand, about 1 000 demarcators were recruited for the demarcation and listing process. Although demarcation was planned to be completed by July 1996, it in fact continued into the enumeration period.

The nine provinces were divided into regional offices (as indicated in Table 12) to provide efficient supervision of the work in progress and to enable those with close knowledge of the area in which they had to work to be recruited.

A team of independent monitors, drawn partly from the advisory committee and partly from demographers and other experts, visited various parts of the country during this process to help ensure that both the demarcation and listings were adequately performed. When the monitoring team found problems in any part of the country, these were attended to. In addition, if the monitoring team was not satisfied with the work that had been done, the demarcation and listing processes were repeated where necessary.

The questionnaire for the census was developed with active participation by the census advisory committee. Items were suggested by various stakeholders, especially parastatals such as Eskom and government departments such as Health, Housing, Education and Welfare. Expert assistance in the design was also provided by the international consultants. It was revised three times, based on comments received, behind-the-glass observations and fieldworker debriefings.

Some questions applied to each individual in the household. For example, respondents were asked to indicate the age, gender, population group, home language, education level, occupation and income of each person present in the household on census night. Other questions dealt with the circumstances of the household as a whole, for example, whether or not the household had access to electricity and piped water. Altogether, there were 73 items of information requested in the questionnaire, 50 applicable to individuals, and 23 to households.

Separate, briefer questionnaires were developed for people in institutions such as tourist hotels, homes for the aged, police cells and prisons, and for homeless people.

The questionnaire was tested in a pilot survey of 50 000 households in May 1996, conducted with the assistance of the Human Sciences Research Council. It took an average of 15 minutes to complete it in urban areas, and 30 minutes in non-urban areas. Few problems were encountered with regard to the questions asked, and how they were phrased. Where problems did occur, as indicated by fieldworkers or inspection of the answers, the questionnaire was modified accordingly. The questionnaire was then finalised and translated with assistance from South African academics into all 11 official languages for use in the full population count in October.

In the pilot survey, it was found that most people were willing to answer all questions, including potentially sensitive ones regarding population group or religion.

During the pilot survey, fieldwork arrangements for the final count were also tested in practice. For example, methods of identifying EA boundaries and visiting points, and planning a route through the EA for questionnaire distribution, came under scrutiny.

A publicity campaign was launched in the mainstream media, focusing initially on radio advertising and talk-shows for non-urban coverage and, in the final stages, on print, television advertising and interviews. The campaign, which was coordinated for Stats SA by a communications company with associates in the nine provinces, employed both an advertising agency and a market research consultancy.

The mainstream media campaign was complemented by a grassroots campaign, including provincial launches, road shows, visits to 300 local authorities, meetings with more than 500 civic and traditional leaders, the distribution of four million Census ’96 comics to school children, and the dissemination of 14 million brochures in all official languages with assistance from churches and community structures.

Partly as a result of the inadequacies of previous censuses with regard to demarcation, especially in the former ‘homelands’ and ‘independent states’, and partly as a result of the limited time available, demarcation was not uniformly completed throughout the country prior to the enumeration phase.

In certain non-urban areas and informal settlements, where maps were out-of-date or unavailable, the intention was to take aerial photographs on which to base demarcation. But, in most of these areas, heavy rains and cloud-cover made aerial photography impossible. Reliance had to be placed instead on descriptions of ground features or lists of households supplied by traditional authorities to determine the boundaries of these EAs, particularly in some non-urban areas in Eastern Cape and KwaZulu-Natal. As a result, boundaries were not always clear.

The enumeration phase made use of 11 provincial offices (three in Eastern Cape and one in each of the other provinces) and the 72 regional offices throughout the country, established during the pre-enumeration phase. These offices had to be staffed and equipped with networked computers to run the administrative systems.

Approximately 86 000 enumerators (one per EA), supervised by 16 000 chief enumerators, in turn managed by 2 000 controllers, 72 regional managers and 11 provincial census managers, had to be appointed and trained.

The positions to be filled were widely advertised in the press. Applicants with a minimum qualification of matriculation were invited to apply for the enumerator posts. The vast majority of those who were given jobs as enumerators were unemployed at the time. It was decided that Some of the supervisory posts, particularly the positions of controller and higher, would be filled by people with post-school qualifications. More than 500 000 people applied for approximately 100 000 posts.

Training manuals were developed for enumerators, chief enumerators and controllers. The training was handled using a cascade approach. Stats SA invited staff from universities and technikons to assist with training throughout the country. Eighty of these volunteers were given training over a period of three days by the census staff in Pretoria. They were then dispersed to the various regional offices to train the controllers, who in turn trained their own chief enumerators and enumerators in each regional office.

After training, each person selected to participate in the work of data collection for the census was expected to sign a contract in which the job requirements and rates of pay were stipulated.

Before counting could start, material had to be distributed through 48 depots to all parts of the country. This included census documents, ID cards, pens, over-shirts and satchels. The questionnaires alone weighed 740 tons. Special arrangements were made with the State Tender Board to improve timeous procurement. Transport and communication arrangements had to be made to enable all parts of the country to be visited, and to ensure that problems in the field could be promptly reported and adequately handled.

A team of six temporary employees was appointed to serve as internal auditors. Its members travelled among the 72 regional census offices, giving administrative training and controlling the cash accounts which were managed from these offices.

Enumeration of approximately nine million households started on 10 October and continued into November 1996, using the night of 9-10 October as a reference point.

The respondents were given a choice regarding the method of questionnaire completion. The enumerator could complete it for them during a face-to-face interview in their language of choice, or else they could opt for self-completion. The vast majority of people chose to be interviewed.

The counting process actually began on the evening of 9 October, when organisations for homeless people countrywide teamed up with census staff on city streets in a pioneering effort to enumerate the homeless.

In addition, questionnaires were administered in various institutions, for example prisons, police cells, tourist hotels and homes for the aged and disabled, using specially trained enumerators.

The overall process of enumeration took longer than anticipated, and continued into November. Delays were caused by disruptions to training instigated by unsuccessful job applicants, poor weather in parts of the country, and the large proportion of people opting to be interviewed, rather than filling in the questionnaire themselves.

Completed questionnaires were taken in EA boxes to the regional managers at the 72 regional offices around the country.

Once a particular visiting point had been enumerated in the EA, the enumerator was required to enter into the census summary book the number of males and females, and the total in each household, at that specific visiting point. He or she also had to indicate all vacant premises or stands and premises which contained non-residential buildings such as churches, schools and businesses, as well as households which refused to co-operate, or else could not be contacted despite repeated calls, in that EA.

Fieldwork information regarding the number of people by gender in each EA was also entered on to a control form eventually signed off by the relevant controller. This form had three purposes:

• It provided evidence that the fieldwork had been successfully completed.

• It initiated the payment process.

• It provided information necessary for the establishment of an administrative database.

During the data-capture phase, it became clear that some of the provincial processing centres had not completed and submitted all their administrative documents at the time that the sample for the preliminary estimates was drawn. This omission influenced preliminary estimates, particularly in relation to drawing a sample of EAs from which to extract questionnaires (see Chapter 3).

These omitted documents were, however, brought into account during the process of capturing them onto computers, and could thus be included into the improved undercount calculations. This increased the final totals.

All census personnel expected to be paid before the start of the festive season. Payment of 80 000 went relatively smoothly. But payment to the remaining 20 000 was delayed as a result of unforeseen problems such as incomplete records, the splitting of an EA between two enumerators during enumeration or the replacement of an enumerator mid-stream during the process, and problems with the computer network. Local management capacity to resolve the problems and operate the payment software was uneven across the 72 offices, leading to a payment crisis.

The payments records of the outstanding staff were moved to head office, and the matter resolved from there over the next two months.

As a result of the payment crisis, the checking of the adequacy of enumeration was uneven in some offices, and the start-up of processing the detailed results was delayed.

Some of the problems experienced during the enumeration phase are listed below:

• Access was difficult in certain of the more-affluent urban areas where many households live behind high walls, electrified fences and security gates. This resulted in a proportion of non-contacts of households.

• Enumeration of those people who are hard-to-reach in any country in the world, such as the homeless, young people living on their own, and the under-five year olds, also presented problems in this country.

• People living on commercial farms in some areas also proved hard to reach, sometimes because the farmers refused to participate.

• Uneven checking procedures influenced the quality of the data gained in some cases.

• Instances of cheating and falsification by a small proportion of enumerators, often using unforeseen methods, became apparent.

• Dealing with the payments crisis meant that some of the regional offices had to be closed and their documents moved to the provincial offices before the process of checking the contents of all the EA boxes could be undertaken.

Coverage problems were taken into account in the post-enumeration survey, and in the adjustments for undercount, as described in Chapter 3.

During the data processing phase, information collected on the questionnaires during the enumeration phase was converted into electronic format. Coding of open-ended questions such as occupation, the actual data capture, checking and processing was undertaken at nine sites, one in each province, throughout 1997 and part of 1998. Five thousand temporary staff worked in three shifts to complete the task.

Preparation for data processing involved:

• Identifying and renting office space in each province, large enough for the installation of all computers, work and storage space.

• Developing coding lists for open-ended questions, and coding manuals covering all questions.

• Creating tabulation plans to ensure that the data could be captured in appropriate ways for analysis.

• Deciding on appropriate software for data capture, storage and back-up.

• Writing computer programmes for data capture, editing and verification.

• Installing the computers linked to a wide-area network.

• Recruiting and training suitable people for both coding and data-capture.

• Developing a work-related payment system.

• Working out work-flow and management methods.

• Collecting all the boxes containing the questionnaires (EA boxes) and bringing them together in each provincial processing office.

• Sorting these boxes and storing them.

Recruitment and training for data-processing staff took place at the various data-processing centres around the country. Data-processing staff were trained by provincial officials, who had previously been trained by staff at head office.

Management structures were established, not only at head office, but also at each provincial centre. At head office, central processes affecting all operations were managed, such as those listed above.

The Stats SA management team at each processing centre consisted of the head of the centre, a chief storekeeper, a chief administrative officer, a personnel officer and other support staff. In addition, a tender was awarded to a management consulting firm in the private sector to assist with managing the processing in each provincial centre.

The exercise of data processing was split into a number of processes, including data coding, data entry, throughputs and methods of outputs.

• During coding, questionnaires had to be carefully checked and the answers correctly coded. The coding manuals gave guidelines for coding all questions, and code lists were provided for open-ended questions such as occupation and economic sector. The coding done in the field had to be verified or, where necessary, altered. Codes were directly filled out on the questionnaire with a clearly visible marker.

• During data entry, the conversion of data from hard copy format into electronic format took place, including editing and data verification.

Data integration involved bringing the data from the provinces to head office, running additional data quality checks and writing additional editing programmes, combination of the data sets, checking on the data quality, and writing programmes for derived variables. A task team was formed to assist with data editing and cleaning and the development of derived variables, consisting of both subject-matter specialists, such as demographers and economists, and computer experts. This team developed suitable range and consistency checks, and gave guidelines on their implementation

There were three main problems experienced during the data-processing stage.

• The first concerned human resources and the use of temporary staff, largely drawn from the unemployed in the provinces, for data coding and entry. Labour unrest in certain centres and complaints about appointment procedures and payment systems caused considerable delays.

• The second involved uneven management of the processing centres in the provinces. For example, some processing centres were better organised in terms of storing and sorting the EA boxes, and faster at processing than others. This meant that, towards the end of processing, boxes had to be transferred from one processing centre to another to enable the work to be completed in a reasonable time period.

• The third consisted of technical problems, for example the network being ‘down’, preventing data-capture on certain days; under-estimation of the computer capacity that would be required, resulting in the need mid-stream for upgrading the hardware to cope with the massive volume of data; the inappropriateness of the database software originally chosen for data editing and the development of derived variables in such a huge dataset; and the interfacing from one system to another to speed up data cleaning.

The development, dissemination and marketing of products is an ongoing process, and will continue for a number of years. It includes deciding on the range of projects to be produced and services to be offered in relation to census outputs; the software to use; the setting up of structures, both at head office and in the provinces, to supply the products or offer the services; and decisions on issuing of free products and pricing procedures.

The overall objective of this phase is to produce and disseminate user-friendly products and services targeted to different audiences. Some background information is given here on the most important products. For more detailed information, the reader is advised to consult the census products brochure. These products can be divided into those that will be available free of charge, and those which will recover some production and service costs. They are described in the box below.

The Interim Statistics Council, a statutory body which advises the Minister responsible for statistics on the quality of statistics throughout the country, set up a census review committee to evaluate the census and its methodology. In addition to four members of the Council, four team members were appointed by the National Economic Development and Labour Council (Nedlac). The names of the people in the review team are indicated in Appendix 1. The report submitted by the task team, on the basis of which the Council advised the Minister in this regard, is available from the secretary of the Council via Stats SA.

The following are some of the possible sources of error in any census (adding to the list by Nel, Loubser and Van Wyk, 1993).

• The boundaries of some EAs are not clearly indicated and, as a result, some parts of the EA may be excluded during enumeration, or else the same visiting point may be enumerated twice.

• The census frame is incomplete in some EAs, since not all visiting points are listed, and the enumerator fails to visit unlisted visiting points.

• The enumerator fails to visit all visiting points, even though they were listed. This occurs as a result of non-contacts, or refusals, or for other reasons (for example failing to follow an appropriate route through the EA, and missing out parts of it).

• The enumerator leaves out a household at a visiting point which contains more than one household.

• Individuals within households are missed.

• The information collected on some individuals is incomplete or inaccurate.

• Coding of the information on the questionnaire is inaccurate.
  

• Numbers are mispunched during computer entry.

People in the following situations during a census are more likely to be missed than others:

• Single people living on their own.
• Those moving around or travelling.
• Farm labourers and others living on farms, where a farm owner refuses to give enumerators access to the property.
• Babies and young children.
• Those in more-affluent areas concerned about security.

The extent to which some of these errors occurred can be identified and taken into account when calculating the extent of undercount or overcount in a post-enumeration survey. In the following chapter, the way in which the 1996 post-enumeration survey was conducted is discussed.