Counting the costs of a twenty-first century census: South Africa's census 2001

Abstract

Increasingly complex societies necessitate the collection of more information, or more sophisticated ways of estimation. This places upward pressure on the cost of collecting such information, as family structures are more complex, mobility more frequent, and willingness to provide information declines. As a result, rising census costs have become an issue in many countries. South Africa's 2001 Census was arguably the most expensive in the country's history. However, assessing the cost of a census requires a closer scrutiny of the cost of census-taking as well as the procedures that were followed and the benefits that might arise. This paper uses information provided by Statistics South Africa to analyse the cost of Census 2001 and explore lessons that may be derived for future census activities.

1. Introduction

While increasingly complex societies necessitate the collection of more information or more sophisticated ways of estimation, the costs of collecting such information are a concern. One aspect of this is the increasing cost of census-taking, which has been experienced in both developing and developed countries (Alam, 2001; Diop, 2001; Leete, 2001; UK Parliament, 2001). Reasons given for the high and rising costs of census-taking include population growth; smaller and more complex family structures; increasing national and trans-national mobility; the high cost of labour-saving and time-saving technologies, and the learning curve associated with changing from long-established methods to new approaches; increased labour costs, and the need to adhere to the employment practices adopted by progressive governments (Simpson, 2001: 423). Other concerns relate to resistance to census-taking by certain populations, attempts to improve coverage of groups traditionally excluded from census-taking, such as the homeless and young males from poor communities, and reliance on private sector suppliers who have access to scarce skills.

The costs and benefits of census-taking, once taken for granted, are now being called into question. How cost-effective a census has been can be thought of as the outcome of a number of inter-related dimensions, of which efficient resource allocation is only one component (Holt & Diamond, 2001: 446).

	First, and most important, is the accuracy of the count and distribution of the population, and the extent to which bias and error can be identified and taken into account.
	Secondly, the breadth and depth of the information that has been collected beyond the population count and basic demographic characteristics of the population.
	Thirdly, the accessibility of the census data to decision-makers and researchers. In part this refers to the timeliness with which information is made available, and in part, to the format that is used when disseminating the results.
	Finally, the way in which the resources made available to the census were deployed to achieve efficient procurement procedures, employment generation, investment in research and development activities and investment in physical capital such as computer equipment. Although this article will largely focus on this component, the success achieved in the three items above must ultimately determine any assessment of a census.

The article starts by examining the lessons that Statistics South Africa1 reported had been learnt from the South African Census held in 1996, for the management of the Census held in 2001. The cost of Census 2001 is then examined in comparison to Census 1996 and the costs of other population censuses in countries similar to South Africa. The article goes on to analyse the items that contributed towards the cost of Census 2001, isolating those components that represent an investment in knowledge or physical capital from those that are the direct and one-off costs associated with a project of this scale. The expected benefits of Census 2001 are then reviewed and finally the article draws out lessons for consideration in planning future census-taking in South Africa as well as broader lessons for the activities of national statistical agencies.

2. Lessons for cost-effectiveness from census 1996

Despite some flaws, Census 1996 is acknowledged as a benchmark for South Africa, as it represents the first attempt to count the population of post-apartheid South African and to obtain essential planning information through the census process. It provided a base against which other data can be placed into perspective, and the R400 million that was spent on it at the time appears to have been a worthwhile investment, despite some concerns over the accuracy of the count.

Several reviews of Census 1996 have been undertaken: these offer important insights that could have facilitated the management of Census 2001. While most of these lessons relate to technical issues such as statistically adjusting for the undercount, improving coverage and ensuring quality control, three issues are important for this review of the costs and benefits of census-taking. These are improvements in organisation efficacy, data processing and the depth of the information that is gathered.

2.1 Organisational efficacy

Administrative capacity was identified as an important constraint in the administration of Census 1996. In 1996, the census project had to create its own administrative support mechanism. This was said to have resulted in duplication and waste through inexperience as well as through deliberate fraud. To resolve this, review documents comment that the Stats SA management intended to develop viable administrative structures that would work in unison with the technical activities (Stats SA, 1999a). In the early planning documents for Census 2001, improvements in administration were identified as a potential strength that would help reduce cost and wastage through effective management and ensure timely delivery of services. Indeed, in July 1999, Stats SA claimed that the administrative wing of Stats SA was probably at full strength (Stats SA, 1999a).

Management capacity was identified as being a related constraint in 1996. By mid-1995 the Central Statistical Service (CSS), the precursor to Stats SA, was still in transition and a new management team had only just been put in place. The internal review of Census 1996 identified that management had been too heavily involved in detail and the ‘big picture’ of the Census was not properly managed (Stats SA, 1999a). In particular, too many new and ambitious projects were embarked on, while logistical planning was poor. In addition, financial management was also identified as being unacceptable. In response to this, early planning documents for Census 2001 indicated that external project managers would be appointed, and that sub-projects would be reviewed with the view of simplifying processes (Stats SA, 1999b).

The final issue reflecting organisational efficacy related to the procurement processes followed in 1996. Considered to be less than satisfactory, the document proposed that the tender process be reviewed to allow fast-tracking, with streamlined processes, in order to overcome problems of delays and inadequate service provision.

2.2 Data processing

Slow and costly data processing was identified as a problem in Census 1996. In this census, about 35 per cent of the data-processing budget was spent on paying wages. Another 15 per cent went on office rentals and equipment. It was judged that excessive time had been devoted to staff management rather than to production. To counter this, the data processing plan envisaged in the assessment of Census 1996 was to use ‘intelligent’ optical reading devices or scanners for Census 2001 (Stats SA, 1999b). It was argued that the initial cost of equipment could be amortised by using this equipment for follow-up household surveys, economic surveys and processing vital data registration. The envisaged plan was expected to reduce the cost of this phase by between 30 and 40 per cent of the cost of Census 1996.

2.3 Information depth

The questionnaire in Census 1996 was long and detailed, and contained 65 questions. This was said to reflect the ambition to produce detailed information about the South African people; however, it was acknowledged that this decision increased the cost of census-taking. It was suggested that planning for Census 2001 should consider concentrating on a few main questions whose answers are essential at a detailed level (Stats SA, 1999a). The length of the questionnaire was also identified as likely to have an impact on almost all phases in census-taking.

3. Has census 2001 cost more?

Given these anticipated improvements, and the hope that was expressed that the costs of the census might be reduced, the next issue to resolve is whether Census 2001 actually cost more than Census 1996, once all of the appropriate adjustments are made for changes in purchasing power and population growth. However, judging the extent to which the costs of the census exceeded expectations is complicated by the existence of numerous budgets and several allocation approvals, not to mention the difficulties associated in estimating cost changes in any census-taking (Edmonston & Schultze, 1995: 44–6).

3.1 Estimating the total cost of census 2001

The following is an incomplete list that provides some idea of the range over which the cost of Census 2001 was estimated (Scott, 2003; Stats SA, nd, Stats SA, nd,a, Stats SA, nd,b, Stats SA, nd,c, Stats SA, nd,d.

	R490,6 million: original allocation, 1998
	R590,8 million: estimate from a Technical Discussion Document prepared in response to the original allocation, 1999
	R626,0 million: original budget prepared in 1998
	R710,6 million: revised allocation, 1999
	R848,3 million: original allocation plus approved additional amount, 2001
	R901,8 million: submission to the Minister of Finance, 2002

The final cost of Census 2001 as used in this article is broken down as follows:2

Table

• Original allocation of	R633 055 000	plus
• Additional approved amount of	R215 275 000	equals
• Total Approved Expenditure	R848 330 000	plus
• Unauthorised over-expenditure of	R122 000 000	equals
• Total Cost	R970 330 000

Based on these figures, an initial comparison of the costs of Census 1996 and Census 2001 suggests an alarming increase in the cost of census-taking, from R408 million to R970 million. Indeed, at one level Census 2001 could be accused of having massively exceeded the original allocation of R490,5 million (in 1999 Rand), while at the very least the Census incurred an unauthorised over-expenditure equal to 14 per cent of the approved allocation.

However, a number of adjustments can legitimately be argued for if a true picture is to be derived of the census costs (Riche, 1997). These include taking account of inflation, population growth and changes in household structure.

After these adjustments Table 1 suggests that the increase in costs between the two Censuses may be less worrisome.3

Table 1:  Per capita cost of post-apartheid censuses

	Census 1996	Census 2001
Real cost	R729 million	R987 million
Population	40,6 million	448 million
Households	9,1 million	11,7 million
Cost per capita	R17.96	R22.03
Cost per household	R80.10	R84.36
Source: Stats SA (nd,a); authors' calculations.

Using estimates corrected for inflation prepared by Stats SA, the cost per person increased from R17.96 in 1996 to R22.03 in 2001, while the cost per household enumerated increased from R80.10 to R84.36. This was caused by a substantial decline in average household size between 1996 and 2001, and hence a greater than expected increase in the number of households that had to be visited. The data in the table thus suggest a real increase in census costs of R258 million, equal to 35 per cent of the total Census 1996 cost expressed in 2001 prices, but only some 5 per cent of the per household cost.

3.2 Allocation of resources

Turning to a comparison of the cost breakdowns of Census 1996 and Census 2001, Table 2 shows comparable activities and their final cost.

Table 2:  Breakdown of activities in post-apartheid censuses

	1996			2001
Activity	R million (current)	Adjusted to 2001 prices	Per cent of total	R million (current)	Per cent of total
Enumeration	222,3	397,2	54,4	577,5	59,8
Demarcation	71,3	127,4	17,5	84,2	8,7
Processing	75,4	134,7	18,5	140,6	14,6
Administration	23,2	41,5	5,7	159,4	16,5
Dissemination	16,1	28,8	3,9	4,4	0,5
Total	408,3	729,5	100,0	966,1	100,0
Source: Scott (2003); Stats SA (nd,a).

These figures show that the total amount, and the proportion, of resources allocated to enumeration increased between 1996 and 2001, while cost savings equal to some R43 million were incurred in the demarcation process. To some extent this represents the investments made in improved mapping and demarcation in 1996 and is hopefully indicative of savings that are possible in future census-taking from such investment in knowledge infrastructure. Savings were also achieved in data processing as a proportion of the total, although it seems likely that some costs may have been categorised differently in 1996 and 2001, given the apparent increase in the proportion of resources expended on administration. Finally, a significantly smaller proportion of expenditure was allocated to dissemination activities in 2001 than had been done in 1996, although this is partly because Census 2001 shows the breakdown of expenditure over the five years of Census activity.

As might be expected, the financial year of the Census (2001/2002) absorbed 76 per cent of all funds utilised over the 5 years in which costs have been incurred. However, when the budget is re-estimated eliminating funds rolled over into later years, the initial expenditure in the first year of the Census process was one tenth the amount budgeted for. This supports an important reason given by Stats SA officials for the problems experienced by the Census: that the initial allocation was insufficient to undertake a census, which delayed planning activities and affected the measures that were to have been implemented to reduce census costs (Stats SA, nd,d).

3.3 Detailed budget breakdown

Table 4 refers to the period from 1999 to the end of the 2003 financial year and indicates the broad categories of expenditure compared to the amounts that had been budgeted in the revised allocation for these categories.4 The table also shows the variance between allocation and actual expenditure.

Table 3:  Breakdown of costs by year

Financial year	R million	Per cent of total expenditure	Per cent allocated by final budget
1999/2000	5,7	0,6	5,3
2000/2001	69,5	7,2	11,5
2001/2002	744,3	76,4	65,6
2002/2003	149,3	15,4	17,6
2003/2004 (est)	5,0	0,6	0.0
Total	973,9	100,0	846,9
Source: Stats SA (nd,a).

Table 4:  Categories of expenditure

Item	Final allocation (R)	Expenditure 1999–2003 (R)	Variance (R)	Variance as per cent of allocation	Per cent of total variance
Personnel	78 522 395	141 876,346	(63 353 951)	80,7	24,1
Administrative	90 731 777	143 447,778	(52 716 002)	58,1	20,0
Inventory	58 353 815	98 015 125	(39 661 310)	68,0	15,1
Equipment	76 905 356	90 729 514	(13 824 158)	18,0	5,2
Buildings	13 459 111	539 863	12 919 248	96,0	− 4,9
Professional services	392 556 646	491 304 364	(98 747 718)	25,2	37,5
Miscellaneous	41 628	3 008 756	(2 967 128)	7 127,7	1,1
Not yet allocated	na	5 000 000	na	na	1,9
Total	710 570 728	973 921 746	(263 351 018)	37,1	100,0
Source: Scott (2003); authors' calculations.

Immediately apparent from this table are the large cost overruns in personnel expenditure (R63 million or 24 per cent of the total overrun), administrative expenditure (R52 million or 20 per cent of the total overrun) and professional services (R98 million or 25 per cent of the total amount of R263 million). While Stats SA management ascribe the first of these items to the extension of the period for enumeration, this level of aggregation conceals the reasons for the cost overrun. Thus ‘transport hire’ and ‘hotel accommodation’ were included as administrative expenses, as was the freight of questionnaires from the USA to South Africa. This also includes the misallocation of some line items. Examples of these include substantial savings in ‘local travel and subsistence’, offset by large cost overruns in ‘hotel accommodation’ and ‘local transport hired’.

To address this problem, all large savings have been reallocated to similar items with cost overruns, with the exception of ‘rentals: land and buildings’, for which no comparable expenditure item could be identified. This results in a total overrun amount of R263 million, with savings in some line items equal to R15,1 million, and an as yet unallocated expenditure of R5 million for 2003/2004. When these are adjusted, twenty items account for 86 per cent of the total overrun and 51 per cent of the total expenditure. The largest overruns occurred in ‘consultancy fees’ (R54 million or 23 per cent of the total overrun), ‘salaries’ (R44 million, 19 per cent of the total overrun), and ‘printing’ (R32 million, 14 per cent). These three items account for over half of the overrun on the allocated amount. Stats SA management have explanations for all three items. The need to bring in outside expertise to assist with the planning of the Census owing to the lateness of the pilot study was the most important, along with the unanticipated problems that occurred with data processing (Stats SA, nd,d). The overrun on salaries arose from extending the enumeration period, while printing costs were extremely high due to unforeseen problems with South African suppliers.

Other items are however problematic, such as the R18 million expenditure on telephones and cell-phones, 35 times more than the R0,5 million that was budgeted for this line item. Overall, a comparison between the detailed allocation and actual expenditure shows that actual expenditure was more than double the budgeted amounts for 53 of the 161 expenditure items, while there was no budget for another 72 items.

Some comparison is possible with Census 1996, and De Beer (2003) estimates that real expenditure on equipment in 2001 prices increased from R33 million in Census 1996 to R98 million for Census 2001, reflecting both the increased cost of capital goods and an investment for Stats SA. Personnel costs also increased from R403 million to R436 million. However, the largest increase was on professional services, which increased six-fold, from R34 million in 1996 to R189 million in 2001.

3.4 Comparison with best practice

International comparison of best practice in census-taking may be of limited value because the environment for the collection of information varies widely, as do issues of the quality and depth of the census (Alam, 2001: 1). In addition to the inherent problems, the period over which Census 2001 was planned and implemented was one of extreme exchange-rate uncertainty, in which the value of the South African Rand fell from ZAR5.53 to the US dollar during the period of project planning to ZAR9.27 when enumeration started in October 2001, reached ZAR13.85 in December 2001 as data processing began, and returned to ZAR10.33 one year later, in 2002 (SARB Quarterly Bulletin, 2002). Nonetheless, a rough attempt to convert the Census 2001 cost in each year to US dollars, using the weighted average of bank daily rates provided by the Reserve Bank, suggests that the Census cost some US$112,3 m. This is US$16,5 m more that the estimate reported by the United Nations Population Fund (UNFPA) in September 2000 (UNFPA, 2000: 50), and gives a per capita cost of $2.51. An even rougher attempt to estimate the cost of Census 1996 in US dollars gives a per capita cost of $2.82 per person at the prevailing exchange rate and prices in 1996, suggesting that Census 2001 may have actually cost less per person in dollar terms than did Census 1996 (authors' calculations). While at the upper end of the scale of census-taking in African countries, this seems comparable to countries such as Mauritius ($2.06) and Zambia ($1.83) (Anon, nd) and is well below figures estimated for Australia ($6.69), the United Kingdom ($5.59) and the USA ($22.25) (Ngwenya, 2002:4). However, it must be remembered that along with Japan, Turkey, Australia and Canada, South Africa has been one of the small number of countries with populations of more than 10 million that undertake a census every five years, thereby doubling the cost census-taking each decade.

4. What were the reasons for the cost overruns?

Reasons for the cost overruns extend across a range of issues, some of which may have been foreseeable. Some relate to the increasing costs of census-taking observed internationally, while others are difficult to account for in a review of this nature and are best dealt with through normal public accounting procedures.

Attempting to explain the cost overrun, Table 5 shows how Stats SA allocated the additional expenditure on activities that should have improved the census count.

Table 5:  Improvements on original planning

Activity	Cost (R million)	Per cent of total
Listing during the pilot	20	8,7
Management of enumeration	100	43,5
Fieldwork information system	30	13,0
Payments for additional fieldwork	16	7,0
Training for census pilot	24	10,4
Processing of census pilot	12	5,2
Printing of questionnaires	23	10,0
Census at schools	5	2,2
Total	230	100,0
Source: Stats SA, (nd,e).

From the perspective of Stats SA, the most important reason for the cost overruns was the management of the enumeration process, which accounted for some 43 per cent of their estimate of the total overrun. This activity was sub-contracted and is thus included as ‘consultancy fees’ in the line item budget. Upgrading the fieldwork information system was another important improvement, which may in part account for the high cost of telephony in the census. The cost of labour is also an important component of the cost overrun; the average real wage paid to enumerators increased by 25 per cent between the 1996 and 2001 censuses, from R1 490 to R1 870, representing a R30,4 million increase in the labour costs of census-taking. Training for the census pilot would be included under the overrun for salaries. Finally, the estimate for printing costs in this table is far lower than what was actually recorded, and this seems attributable to the exclusion of an amount of R37,9 million in 2002/2003.

However, although these actions may well account for at least part of the overrun, the reasons given for taking these decisions need to be discussed in more detail.

4.1 Inadequate initial allocation

The amount originally allocated to Census 2001 by the then Department of State Expenditure was R490,6 million, R139,5 million (20 per cent) less than the amount originally budgeted. After adjusting for price and population increases, this implies that Census 2001 would have had to be completed at a per capita cost some 40 per cent lower than that achieved by Census 1996. The response of Stats SA to this was to convene a technical discussion group to explore possible alternatives to the taking of a complete census, such as the collection of a 15 per cent sample survey. The estimates shown in Table 6 compared the cost of a census to the cost of conducting a survey that would sample 15 per cent of South African households.

Table 6:  Comparative costs of census and survey

Activity	Estimate for 2001 Census (R million)	Per cent of total cost	Estimate for 15 per cent sample survey (R million)	Per cent of total cost	Sample survey as percentage of 2001 Census
Demarcation	111,2	18,8	111,2	37,1	100
Enumeration	305,5	51,7	101,8	33,9	30
Data processing	95,6	16,2	47,8	15,9	50
Admin. support	27,5	4,7	13,7	4,6	50
Publicity	36,0	6,1	18,0	6,0	50
Dissemination	15,0	2,5	7,5	2,5	50
Total	590,8	100,0	300,1	100,0	50,8
Source: Stats SA (1999b).

Although the sample survey did allow the possibility of cutting costs by half, to be achieved largely through effecting savings in the enumeration phase, it was decided after consultation with the Statistics Council and the Minister of Finance to proceed with Census 2001 (Stats SA, 1999b). The most immediate effect of this was a delay in the pilot study from October 2000 to April 2001, with the consequence that the pilot study results could not be absorbed before detailed planning for the Census commenced.

Finally, it is difficult to estimate the impact of the currency fluctuations on the cost of the Census. While a substantial proportion of the Census cost overrun was on personnel and administrative expenses, the largest item was on consultancy fees. In many instances, these fees are linked to the international rates that are paid to professional organisations, many of which are multinational corporations. It is possible that some portion of these costs can be attributed to the currency fluctuations. Certainly, it has to be recognised that this was an uncertain period in which the planning of large projects such as a census was made more complex.

4.2 Late pilot study

The late pilot study has repeatedly been given by the management of Stats SA as an important reason for the decisions taken to modify the census process as shown above (Stats SA, 2002). A pilot census is a critical test which informs census management of the resource requirements and steps needed to ensure a successful population census. International best practice suggests that a pilot should be run between 12 and 18 months before the full census, thereby allowing sufficient time to implement the lessons and shifts in direction, including the budgetary recalculations, indicated by the outcomes of the pilot (Strengell, nd).

The pilot for Census 2001 was run in March 2001, less than seven months before the full enumeration scheduled for October 2001. The key lessons from the pilot began emerging ten days after field operations were concluded, before pilot questionnaires were processed. Management, in anticipation of the impact of these lessons, immediately began implementing the remedial actions listed above to resolve key deficiencies as it could not wait for the detailed report that would only be tabled in September, a month before enumeration was to begin.

Briefing the Statistics Council in April 2001, Stats SA managers concluded that the Census management was in crisis mode well before the actual census-taking process, and noted the following main concerns:

	No integration existed between the sub-projects of the census.
	The critical path required to implement the census was not being respected and deadlines were not adhered to nor regarded as priorities.
	The top management of Stats SA was in disarray, with inadequate internal communication and personality clashes.

Rather than delay the census, Stats SA announced that they intended implementing a Nerve Centre which would ‘…develop, maintain and manage the big picture’. The objective of the Nerve Centre was to bring together management and project managers to make quick but informed decisions and to consolidate status reporting and progress reporting. The Nerve Centre was also expected to consolidate communication between the various individuals and groups. Central to the Nerve Centre was the decision to source external expertise to provide us with additional management capacity. This was largely through the appointment of consultants as project management specialists.

A number of other concerns were also identified:

	deficiencies in the budget and also with expenditure control
	inefficiencies with payments procedures, especially slow turnaround and access to funds.
	inadequate audit trail processes and support systems.

In an attempt to overcome these problems it was proposed that the relationship between financial management of the Census and the financial management of Stats SA as a whole required clarification and that state accountants be appointed in each province.

The pilot study also showed that ratio of one supervisor to every ten enumerators had to be revised to a ratio of one supervisor to every five enumerators. Assessment of the quality of enumeration in the pilot census made it clear that increased supervision, for a longer period of time, was essential to ensure an improvement in the quality of enumeration, and hence in the quality of final census data.

Finally, it is worth noting that the pilot census document makes no mention of the project costs of the census, although there is a heading on timelines and costs. Thus although the pilot study did include data processing, severe problems that were later encountered with the optical scanning activities were not identified, and data capture is barely mentioned in the pilot study report.

4.3 Extended enumeration

During the enumeration phase it became apparent that the time allocated for Census 2001 would have to be extended if the count was to be more complete. This had an obvious impact upon the costs of salaries, as well as a more indirect impact on the costs of transportation and administration. Although the extension should have minimised bias in the Census, there were problems with the way in which these mop-up activities were undertaken. In particular, because the mop-up was not anticipated, the logistics seem not have been properly planned, monitored or documented. This problem was exacerbated when the field workers went on strike because the payment sub-contractor failed to deliver.

The improved post-enumeration survey is another reason given by Stats SA for higher costs in Census 2001. The differences in the amounts allocated for this activity is shown in Table 7.

Table 7:  Post-enumeration survey

Census 1996 (R million)			Census 2001 (R million)
Expenditure	Budget	Variance	Expenditure	Budget	Variance
3,3	0,0	(3,3)	9,5	3,9	(5,6)
Source: Scott (2002).

No allowance was made for a post-enumeration survey in the original Census 1996 budget. This oversight was corrected in Census 2001, when the amount that was allocated was just over what had actually been spent in Census 1996. Had price changes been taken into account, this budget should have been in the order of R5,9 million, assuming that no cost savings could be identified. Even so, the final cost of the PES amounted to almost R4 million more than this, with Stats SA attributing this cost overrun to the thoroughness of the reconciliation between the Census and PES records, and repeat visits to confirm boundaries and improve quality control (Stats SA, 2002). This suggests that the post-enumeration survey should be a better instrument for the statistical adjustment of the Census in 2001 than it could be in Census 1996, and thus should also be a better indicator of the census undercount.

4.4 Management and administration

Management problems compounded the problems of achieving a cost-effective census. This started with the budgets prepared for Census 2001. As has been shown, the budget originally prepared by Stats SA, and the revision that was eventually allocated, diverge significantly from the actual expenditures. To some extent this may have been the result of an unrealistic attempt to accommodate the cost limits expected by State Expenditure, but it also suggests that the original planning for the Census may not been fully thought through. Subsequent efforts to estimate a more realistic costing appeared to have lacked the information necessary for estimating a proper allocation of the resources. As a result, it appears that the budget was not used as a planning aid, and the goal of better financial management, which had been identified in the review of Census 1996, was not met.

The Census Administration System (CAS) was also been identified as a problem. This data system identified enumerators down to a supervisor level, rather than to an enumerator area level. This does not permit problems to be identified with data collection at the appropriate unit of analysis, that is, at the point of collection. Instead, it is argued that a CAS system will be needed, at least at the regional level, to provide this information. This would show whether an enumerator had gone to the area allocated to that person; whether the correct bar-coded questionnaires had been distributed and returned; and whether enumeration and returns were proceeding as planned.

4.5 Procurement and consultant support

Despite the hoped-for improvements in administrative and project management capacity, the pilot study highlighted the extreme fragility of both systems. Eventually resort was made to consultants in the expectation that private sector efficiency would resolve the problems. However, at R164 million, consultancy fees are the single largest variance that diverges from any of the budgets prepared by Stats SA or approved by Treasury. They thus require additional explanation. The responsibilities were spread between the main suppliers as follows (Stats SA, nd,a):

Table

• Project management	18 per cent
• Remuneration of enumerators:	8 per cent
• Processing of questionnaires	59 per cent
• Publicity	15 per cent.

Project management consultants were contracted to develop a census administration system (CAS), which included management of the warehouse, packing, receiving materials in each regional office, handling appointments, allocating people to areas at a cost said to be R30 m. Human resource consultants were also subcontracted to manage payments using the CAS and paid R16 million for this. The project managers were also to ensure that accurate records of all financial transactions were kept, including subcontracts. Information technology consultants were to take responsibility for the design of the ‘intelligent’ automation system and the subsequent data processing on this system.

The initial tender document for data processing was clear about the tasks expected of the successful supplier. These were:

	check-in and tracking of census questionnaires
	the efficient extraction of data from Census 2001 and Post-Enumeration Survey questionnaires and images
	efficient operation of data capture and quality assurance procedures, within defined intermediate processing milestones for zero backlog, and predictable time frames
	ease of technology upgrade, compatibility with modern software and hardware designs used by Stats SA
	remote and automated access to the major subsystems for data processing which would permit status monitoring, quality control and performance management.

The data processing problems that actually occurred included incorrect identification of the hardware required to complete the task, inappropriate tracking and data quality procedures and the use of proprietary software that Stats SA were unable to adapt to meet their specific needs. These problems placed unexpected pressure upon Stats SA managers and technicians, who had to work with the consultants to find solutions to the scanning and validation problems. This represents a substantial and hidden cost of Census 2001 that also detracted from the other activities of Stats SA.

A number of other procurement problems were identified of which the most costly related to the printing of the census questionnaire. Timely delivery of questionnaires, in adequate numbers, and to the quality appropriate for scanning technologies, was a critical deliverable for the success of Census 2001. When it emerged that the only local supplier able to print questionnaires to specification was falling behind schedule, and that there were questions over print quality, it became necessary to seek and alternative for the balance of questionnaire printing. In addition, when the finalisation of the GIS project indicated that the number of enumerator areas defined had increased from the projected 70 000 to 81 000, it became necessary to increase the print order for questionnaires. The financial implications of this involved both the increased costs of printing (R30 million) and the cost of transporting questionnaires from the United States, where a supplier that could meet these criteria could be sourced, to South Africa (R10 million). The overall financial implication estimated by Stats SA was therefore R40 million.

Having discussed and compared the cost of Census 2001, and reviewed the reasons given for the significant cost overrun, it remains to briefly consider the benefits of the census-taking.

5. What are the benefits of the census?

In addition to the value generated by census-taking when put to use for decision making, planning and the allocation of public and private sector resources, benefits flowing from census-taking include the immediate impact of the jobs that are created and equipment that is acquired, the research and development activities made during and after the census process which constitute an investment in ‘knowledge capital’.

5.1 Planning and resource allocation

Counting the population is important for planning in its own right, especially in the context of the system of provincial governance that characterises South Africa. The most obvious reason for undertaken a census is then for improved planning and resource allocation through the formulation and evaluation of policies and interventions. This not only affects public sector resources since the private sector also makes extensive use of the census. This is the major contribution of most census that are undertaken, and in the United Kingdom, it is estimated that the census costs less than one twentieth of one per cent of the total amount of public resources allocated over a ten-year period on the evidence that it provides (UK Parliament, 2001). Other planning benefits that rely upon an accurate census include elections planning such as the demarcation for elections, and the multi-billion private sector advertising sector. Census data are also used as the dominator in most monitoring indicators used in assessing coverage or the attainment of targets.

The allocation of resources between provinces is one of the most important aspects of this usage given the structure of federal government in South Africa, and amounts to estimated R600 billion over the five-year period between census years (Republic of South Africa, 2003 and authors' calculations). Viewed in this way, the cost of the census is less than one-sixth of one per cent of the value of just one component of public sector allocation that relies upon an accurate count for their allocation. Calculated over a ten-year period, this percentage seems comparable to that estimated by the UK National Statistician to the UK Parliament.

Maximising the return on a census requires more than simply the use of the population count in the allocation of national revenue and includes the other information that is collected during enumeration. In this regard, the South African Census 2001 was a comparative lengthy questionnaire comprising questions on 45 population items, and 75 questions in total. This needs to be compared to Census 1996 (65 questions), the 2001 Census in the United Kingdom (41 questions), and to censuses in Kenya (20 population items), Zambia (30 population items) and Tanzania (23 population items) (Anon, nd, 3).

Accurate and accessible census results and adequate information with which these results can be projected between census years are thus central to proper planning and resource allocation in South Africa. Estimating the benefits that result is difficult since the cost of error is enormous both in financially and politically. Even recognising that any census result will contain a margin of error, even flawed information allows for more objective and transparent allocation of public resources if there is agreement over the extent and distribution of error. Wide dissemination is required through which greater awareness of Census products can be achieved, along with greater usage of the different results with hopefully an enhanced impact on the quality of life of South Africans.

5.2 Employment creation

Although not a direct intention of most census-taking, job creation has been catered for in both of the post-apartheid censuses. By trying to include as many educated but unemployed people into demarcation, enumeration and processing activities, this aspect of the Census can be thought of as a large scale public works programme that has the potential to create jobs in the services sector. Table 8 shows that over 4,5 million person-days of work were created in short-term contract appointments during Census 2001, equal to just over 17 000 jobs (person-years of work).

Table 8:  Employment creation

Census 2001	Persons	Time (days)	Person-days	Jobs (one year of work)
Listing	16 976	66	1 120 416	4 244
Enumeration	83 123	22	1 828 706	6 927
Supervision	16 136	66	1 064 976	4 034
Controllers	1 200	88	105 600	400
Data capturers	1 400	396	554 400	2 100
Total			4 674 098	17 705
Valued as a public works programme				R265 600 000
Source: Stats SA (nd,a).

Using the conventional public works programme, it can be estimated that job creation of this scale would have cost R265,6 million.5 This is an improvement on the employment creation achieved during Census 1996 which can be estimated to have created 11 000 jobs that would have required R159 million in 2001 prices, if valued as a public works programme.

5.3 Knowledge infrastructure

The technical, administrative and physical capacity of the census-taking organisation is an important issue that has to be taken into consideration when assessing the investment component of Census 2001 costs. This improvement in South Africa's knowledge infrastructure includes investments made in new equipment and procedures, electronic information systems and the skills of Stats SA staff. The aerial photography, demarcation and GIS activities undertaken for the census are another potentially useful resource to the Independent Electoral Commission, as were these activities in Census 1996.

The automated geographic dictionary that has been established using these data is an important investment for future census work, other surveys, and potentially the activities of other government departments that will be brought into a national statistical system over time. In addition, electronic dictionaries on occupations and industry have been established along with the intelligent electronic recognition procedures required to build further dictionaries. Finally the equipment, procedures and programmes are in place for future censuses or other data collection activities. Indeed, Stats SA management are of the opinion that only Australia has reached South Africa's level of development in electronic dictionaries for census data processing (Stats SA, 2002).

Other important longer-term investments took place in the data processing stage of the census, including scanners, digital and physical storage capacity, a local area network, image processing equipment and software and an operational large-scale data processing centre. Many of these investments can be directly carried over and used in surveys and if properly maintained, all of these will be usable for future census activities. This is not to say that data processing for a future census will be cost-free, as additional software costs will be incurred, maintenance, consulting and staff fees will have to be paid, some modifications will inevitably be necessary to accommodate the new questionnaire(s), and so forth.

5.4 Research and development

The research and development activities undertaken for the census may be thought of as in investment in knowledge capital. This refers to investments made in technical know-how both in the areas of demography and information technology, as well as in project management skills that can be applied to other large-scale data gathering projects and forthcoming census work. It is difficult to estimate the value of this type of investment, and much will depend upon the extent to which Stats SA is able to retain the expertise developed during Census 2001, and the willingness and ability of the organisation to learn from the enterprise.

6. Conclusion: Implications for Future Census-Taking

Census 2001 represents a tremendous investment for South Africa, costing as much as one year of pension payments to 100 000 elderly people or a Basic Income Grant to 700 000 people. This R1 billion labour-intensive project, which had to be completed within 5 years, is among the most ambitious enterprises undertaken by the South African government. However, when properly used, the contribution made by an accurate count of South Africa's population to better evidence-based planning and the allocation of public and private sector resources would clearly outweigh the opportunities foregone by using these resources elsewhere. Census-taking should thus be given the resources that are necessary to ensure success, and there should be no uncertainty as to whether and when sufficient resources will actually be made available. Nonetheless, as Ngwenya (2002: 4) has asked, ‘…does adequate funding guarantee good census results?’ In reviewing international evidence, and using undercount as a measure of success, he concludes that although the evidence is mixed, ‘…it is critically important that adequate funding for such a massive and complex operation be secured’.

The evaluation of Census 2001 conducted by the Statistics Council concluded that the count was adequate for most planning purposes, although the final figures concealed a number of compensating distortions in the composition of gender, age and population group. In fact, a special warning was given concerning the accuracy of the labour market information (Statistics Council, 2003). An under-count of 16 per cent eventually emerged from the census, which is well above what is deemed acceptable by Ngwenya (2001).

That unauthorised over-expenditure of R122 million occurred is of concern, as is the fact that Census 2001 cost at least R50 million (7 per cent) more than Census 1996, even when all possible adjustments are made to allow for price increases, changes in household structure and population growth. Perhaps even more worrying is why, in its own budget, Stats SA initially anticipated that the cost per person of the count would decline from R18 to just under R14, without making any adjustment for price changes, at a time when international evidence had overwhelmingly indicated that the costs of census-taking had increased (Alam, 2001; Leete, 2001; UK Parliament, 2001). While some of this has to do with inadequate planning in the initial budgeting of the Census, the ethos of tight fiscal discipline that prevailed in the late 1990s is also likely to have played a part. It is now hoped that down-sizing, to the detriment of right-sizing, has been recognised as a hindrance to the production of accurate and timely official statistics. Moreover, the experience of the two post-apartheid censuses should mean that better information is now available for future planning.

In two areas at least, the cost overruns of Census 2001 are of concern. The first relates to the extremely high costs of communication and co-ordination. The second relates to the reliance upon costly private sector suppliers (R164 million, equal to 17 per cent of the total cost of the census), and what seems to be largely disappointing delivery by some of these suppliers. Such reliance will need to be carefully reviewed in future. Census planning, just as census-taking, may well be one of the activities best left in the hands of the official statistics agency supported by its own skilled project managers.

Another concern is that not all the learning gained from Census 1996 seems actually to have been absorbed. The events of the last decade of apartheid government, and the tremendous changes that occurred during the 1990s, meant that census-taking in 1996 required new institutional capacities. This should have created some form of basis on which Census 2001 could have been built, even though there was a high turnover of staff between the two Censuses. However, administrative systems were still inadequate, senior officials remained deeply involved in detailed planning and day-to-day management, the procurement of services was again a problem, and the breadth of information collected by the Census again introduced additional complexity. The suggestion of using ‘intelligent’ automation for data processing was taken up in Census 2001, although the anticipated cost savings did not emerge. All of the above were compounded by the initial uncertainty over the resources that were available. Yet the wide divergence between the planned expenditure and actual expenditure suggests that these problems would still have emerged, even if the Department of State Expenditure had agreed to the initial budget. Solving the problem of planning, learning from Census 2001, and ensuring that the skills to manage a public enterprise of this scale are in place, must be given high priority if future census-taking is to be cost-effective and the considerable investment in physical and knowledge capital is to be realised.

If savings are to be achieved in future census-taking, it will be important to know how to contain the costs of census management activities. In 2001 some of the costs incurred may have been worth while, in that consultants were successfully used to manage the distribution and collection of the census forms. Other costs, such as those for payment and recruitment of staff, and for the design of the data-processing procedures, seem to have been less well handled. Finally, the management tasks involved in a census must not be underestimated, and these call for far more than technical or statistical skills. Aptly compared to a vast oil tanker that once set in motion can only stop when the data are finally released, census-taking is no place for crisis management.

The analysis in this paper and the conclusions that are drawn are those of the authors only, and do not reflect the views of the Statistics Council or the University of Natal. The information and assistance provided by Statistics South Africa staff and management are however acknowledged, as are the comments received from the members of the Census sub-committee of the Statistics Council and two anonymous referees. An earlier version of this paper was presented at the Joint Population Conference on Urban and Rural Sustainable Development, 13–17 October, 2003, Potchefstroom.

Additional information

Notes on contributors

Pali Lehohla

Associate Professor, School of Development Studies, University of KwaZulu-Natal; and Statistician-General, Statistics South Africa

Notes

¹Hereafter referred to as Stats SA.

²A spreadsheet provided by Stats SA will be used for analysing the breakdown of expenditure in comparison to the budgeted amounts (Scott, 2003; Stats SA, nd,a). Thus all analysis at the line item level sums to R968,9 million. During the time of the Census planning and implementation, the South African Rand experienced substantial fluctuation against most major currencies, although at the time of writing the exchange rate is approximately R6:$1.

³The adjusted population and household figures have been used. Prices were corrected by Stats SA, using the Consumer Price Index (De Beer, 2003). This is a still a somewhat crude estimate, as the prices of different components of the Census would have increased at different rates, and a substantial proportion of Census costs were absorbed by wages and the cost of hiring professional services. In addition, the currency fluctuations over the period of the census would have affected the cost of imported goods and services. Finally, as De Beer comments, in Census 1996 Stats SA administered an amount of R100 million for aerial photography on behalf of the Independent Electoral Commission. This amount was included as part of the main budget in Census 2001, and it is not clear whether the amount has been allocated to the costs.

⁴It must be noted that the approved allocation was actually R848,3 million and that this table does not represent unapproved over-expenditure. However, a detailed budget for this amount was not available.

⁵Estimate based on public works wages plus administrative cost taken from McCord (2002).