PSC. Research Reports. Population Studies Center. Carol E. Kaufman South African Demographic and Health Survey: Methodology and Data Quality

Transcription

1 Carol E. Kaufman South African Demographic and Health Survey: Methodology and Data Quality Report No Research Reports PSC Population Studies Center University of Michigan

2 The Population Studies Center at the University of Michigan is one of the oldest population centers in the United States. Established in 1961 with a grant from the Ford Foundation, the Center has a rich history as the main workplace for an interdisciplinary community of scholars in the field of population studies. Today the Center is supported by a Population Research Center Core Grant from the National Institute of Child Health and Human Development (NICHD) as well as by the University of Michigan, the National Institute on Aging, the Hewlett Foundation, and the Mellon Foundation. PSC Research Reports are prepublication working papers that report on current demographic research conducted by PSC associates and affiliates. The papers are written by the researcher(s) for timely dissemination of their findings and are often later submitted for publication in scholarly journals. The PSC Research Report Series was begun in 1981 and is organized chronologically. Copyrights are held by the authors. Readers may freely quote from, copy, and distribute this work as long as the copyright holder and PSC are properly acknowledged and the original work is not altered. PSC Publications Population Studies Center, University of Michigan S. University, Ann Arbor, MI USA

3 South African Demographic and Health Survey: Methodology and Data Quality by Carol E. Kaufman Research Report No June 1997 Abstract: The South African Demographic and Health Survey (SADHS), conducted by the Human Sciences Research Council (HSRC) of South Africa, is the only national study of South Africa which recorded detailed fertility, health and mortality experiences of almost 22,000 reproductive aged women in the late apartheid era. Political circumstances inhibited the dissemination of the data, and minimal effort went towards the documentation of the survey design, sampling, or fieldwork. Drawing on fieldwork reports, sampling schedules and interviews, this paper documents the methodology of the survey and assesses the quality of the data. The study was conducted under less than ideal circumstances: the HSRC s association with the government evoked a degree of suspicion; researchers were subject to international isolation; and many areas experienced high levels of political unrest. The study did not use identical questionnaires across geographical areas and population groups, questionnaires were not translated into the major language groups, and sample design, training, and fieldwork were not properly documented. However, the investigation shows that to the extent possible, sample design and interviewing techniques accommodated difficult conditions of sampling and fieldwork and do not introduce substantial bias into the sample. Quality checks show that distributions of characteristics of each race group sample are consistent with expectation; black urban samples are probably wealthier and more established than the true population; substitution of non-sampled households probably occurred in all samples, although evidence suggests substitution was minimal; and most data quality checks showed patterns similar to those found for other southern African countries participating in the USAID/ Macro DHS surveys. Dataset used: South African Demographic and Health Survey (SADHS): South Africa, The Author Carol E. Kaufman, Berelson Fellow, Population Council, Policy Research Division, One Dag Hammarskjold Plaza, 9 th Floor, New York, NY Acknowledgments This paper benefited greatly from the co-operation of many researchers and fieldworkers currently and formerly affiliated with the Human Sciences Research Council (HSRC). The report and analyses would not have been possible without the detailed information provided by Dr. WP Mostert, principle investigator of the study, and by Prof. DJ Stoker, the statistician responsible for the sampling, and the many hours of conversation and discussion of survey work at the HSRC with Mr. Johan van Zyl, research scientist in the demographic analysis unit at the HSRC. This research was supported by an International Predissertation Fellowship from the Social Science Research Council and the American Council of Learned Societies with funds provided by the Ford Foundation. I am also grateful to Ronald Freedman and Miriam King for helpful comments on earlier drafts of this paper. The author takes full responsibility for any omissions or errors in the paper.

4 .

5 The South African Demographic and Health Survey (SADHS) was conducted from 1987 to 1989 by the Human Sciences Research Council (HSRC) of South Africa. This survey is the only national demographic study of South Africa recording fertility, mortality, and health experiences of almost 22,000 reproductive-aged women in the late apartheid era. It is unique in that it surveyed women from all four race groups 1 and all areas of South Africa, including the ten so-called homelands, at a time when apartheid boundaries were still entrenched. Other studies carried out at that time commonly excluded homelands or certain race groups for political or logistical reasons. Indeed, the survey reflects the geo-political structures of South Africa. It is a pooled set of fourteen surveys, one for each homeland, and one for each race group in the White Republic of South Africa (RSA). As such, the survey provides important information about demographic processes throughout South Africa at a time for which few other sources of demographic data exist. It is also a study that has been the center of some controversy. The SADHS was modeled on the USAID/Macro DHS studies carried out in other parts of the world, but because of academic and political sanctions the HSRC received no financial or technical assistance from the international demographic community. Political circumstances inhibited the dissemination of the data, and in general, minimal effort went towards the documentation of the survey design, sampling, or fieldwork. As a result, these data have been subject to skepticism from researchers outside the HSRC, both South African and non-south African (Caldwell and Caldwell, 1993). This paper supplements the preliminary background descriptions of the study (Mostert, no date; van Zyl, no date) and addresses some of the specific methodological concerns surrounding these data. While the study design and fieldwork records are no longer complete (some eight years after the study), the following discussion documents to the extent possible the overall methodology of the SADHS and addresses issues of data quality that commonly arise in surveys conducted under less than ideal conditions. An evaluation of the SADHS would be greatly facilitated by contrasting the procedures adopted or the obstacles encountered in this survey with those in studies conducted in other settings However, few surveys are subject to this level of inspection, and comparisons to other studies, beyond basic measures of data quality, is limited. Finally, it should be noted that although I was not involved in any capacity in the design or implementation of the study, I have spent considerable time reading field reports and supervisor notes, assessing the design of the sample and the achieved sample sizes and distributions, and interviewing key players involved in the survey. This paper adopts the format for methodology description used in the USAID/Macro International DHS country study reports. It first provides a brief discussion of the South African geo-political environment, especially those aspects which are pertinent to methodological aspects of the study 2, and is followed by a general overview of the methodology, including sample design, questionnaire construction, fieldwork, and response rates. At the end of this section, specific problems in fieldwork are addressed including substitution, political unrest and suspicion. When available and relevant, details on the methodology of each of the fourteen surveys are included. Finally, the results of analyses investigating the quality of the data are presented. The paper concludes with summarizing comments on the weaknesses, strengths and possible biases of these data. Background, Context, and Terminology At the time of the SADHS survey, South Africa was still operating under an apartheid system in which all South Africans were officially categorized as White, Coloured (of mixed heritage), Asian (or Indian) or Black (or African). 1 Under apartheid, all South Africans were officially categorized as White, Coloured (of mixed heritage), Indian (or Asian) or Black (or African). These classifications were legally sanctioned by the South African government and have been used throughout official statistics and much research on South Africa. Because of this and because persons were indeed treated a particular way socially, legally and economically on the basis of their racial status, this report also uses these terms. I will use African and Black interchangeably to refer to the indigenous population. 2 It is beyond the scope of this paper to address in any detail the system of apartheid and its implications for those living under it. For a more comprehensive treatment, see Kuzwayo (1985), Marks and Trapido (1987), Wilson and Ramphele (1989), and Worden (1994), for a few examples of an extensive literature. 1

6 2 Underlying this differentiation, the ideology of separate development held that these races should develop independently. For each group, the government designated areas in which to reside and imposed restrictions on virtually every aspect of life, including mobility, employment, and educational opportunities. Geographically, for Whites, Coloureds and Indians, segregated residential areas within the Republic of South Africa, were designated as the White area with ten areas or homelands set aside for Africans, ostensibly based on ethnic groupings. Four of these, Transkei, Bophuthatswana, Venda, and Ciskei, eventually became so-called independent states with sovereign rights. The remaining six homelands, KwaZulu, KwaNdebele, QwaQwa, KaNgwane, Gazankulu, and Lebowa, remained a part of the Republic of South Africa, but with self-governing rights within their borders (see map). Many Africans were resettled to these areas with or without their consent, a policy that was especially enforced through the 1970s and into the early 1980s (Platzky and Walker, 1985). However, many Africans remained in the so-called White areas of South Africa in townships 3 and squatter areas for better access to jobs, education, and social services. (According to the 1991 census, 69% of Blacks residing in White RSA lived in urban or semi-urban areas.) This complex system has produced confusing and ambiguous terminology throughout South African research. For purposes of classification here, Republic of South Africa (RSA) or White area refers to all of South Africa exclusive of the ten homelands, South Africa refers to the entire country inclusive of the homelands and independent states, the term homelands means all ten homelands regardless of the ostensible level of autonomy, SGTs refers to the six self-governing territories, and the TBVC states indicates the four nominally independent states (Transkei, Bophuthatswana, Venda, and Ciskei). Although the government had begun to dismantle many social and economic apartheid structures, in the 1980 s most of these raciallynbased geographical designations remained. Table 1 provides population estimates for all ten homelands and for each race group in the White RSA. Estimates are based on the adjusted 1991 census and analogous censuses in the four independent states. Although the adjusted figures most likely underestimate the population, they provide some indication of the overall population distribution of each race group and political-geographic units that comprised South Africa, each of which served as the basis for the fourteen surveys of the SADHS.. The SADHS: Background The geo-political divisions in South Africa and an inadequate registration system for vital statistics especially for Blacks, resulted in incomplete sources of demographic data. In the mid-1980 s, the HSRC 4 suggested to various government bodies that conducting a survey modeled on the USAID Demographic and Health Surveys could provide key information on health, demographic measures and their correlates. The Department of National Health and Population Development agreed that a research program was central to planning for demographic change in South Africa and decided to support the study. Because the ten homelands fell outside the official area of responsibility of the Department of National Health, however, each homeland government had to agree to the study and appropriate sources of funding had to be secured for fieldwork in these areas 5. While many homeland leaders expressed concern 3 Townships refer to urban areas, usually outside of major cities and towns, set aside for non-whites under apartheid. 4 Since the 1960s, the Human Sciences Research Council has conducted the bulk of demographic research in South Africa (see van Zyl, 1994, for a review of this work). This quasi-state research organization had a dubious role in many projects carried out on behalf of the Nationalist apartheid regime, a history it is scrambling to overcome in post-apartheid South Africa. Its background has contributed to the suspicion surrounding demographic research in South Africa, a suspicion which has been exacerbated by the HSRC's prior reluctance to make data and methodological detail available to other researchers. 5 Under apartheid, each of the ten homelands was responsible for its own health system, family planning services, and the monitoring and evaluation of these programmes (Republic of South Africa, Department of Health, 1973, 1975).

7 3 over the highly personal content of the survey, they all agreed that the information was necessary (Mostert, 1994a). The HSRC eventually obtained clearances for all homeland areas with the proviso that each would be provided with a report of the findings after completion of the survey. Funding was provided by the Department of Development Aid for the six self-governing territories, and, for the TBVC states, the surveys were funded through the Department of Foreign Affairs. Goals and Objectives The primary goal of the SADHS was to provide data on fertility, contraceptive use, fertility preferences, childhood mortality, utilization of pre-natal and child health services, breastfeeding practices and nuptiality patterns throughout South Africa. The questionnaire used in the study also collected additional information on women s work history, housing and sanitation and educational attainment. The study was intended as a source of demographic data for use by health departments, planners and policy makers. It was also to update and elaborate upon previous research on fertility and child health. The objectives of the survey were to: assess the overall demographic situation in all geo-political units of South Africa, assist the Department of National Health and the Department of Population Development in the evaluation of their programs and policies for the RSA, assist the governments of the homelands in the assessment of their health programs and population policies, measure the changes in fertility and contraceptive use, and study the factors which influence these changes, examine the patterns and correlates of child health and mortality in all areas of South Africa. The SADHS was not designed as a single survey, but consisted of a series of 14 surveys based on race and geographical boundaries. Each of the ten homelands and the four population groups residing in the White RSA were surveyed independently, although a similar interview instrument was used for each survey. While this complicates methodological documentation, assessment and analysis, such a design accommodated the geo-political circumstances of South Africa. It also allowed for independent analysis for each geo-political area or race group (henceforth the fourteen survey types will be referred to as survey areas ) since the samples sizes for each were sufficiently large to capture rare events. The HSRC also conducted a similar DHS-type survey in Namibia, but those results are not included in this report. Sampling Methodology According to Professor Mostert (1994), the principle investigator, and Professor Stoker (1994), the statistician in charge of the sampling for the SADHS, samples were drawn in much the same manner across survey areas. The identification of households in the field, however, varied considerably across survey areas because of the differences in settlement patterns and the level of detail available in the sample frame. The following sections describe the general approach used in drawing the samples. (Please see the appendix for the variations in these procedures across survey areas.) 1. Target Population and Sample Sizes The target population for the surveys was defined as women between the ages of 12 and 49 inclusive who were currently in a union (living with someone), had been in a union previously, had ever given birth to a child, or who

8 4 were currently pregnant 6. This definition was used because marriage in many groups in South Africa is processual and sexual relations occur between couples who may not consider themselves traditionally or legally married. Asking only women who considered themselves ever-married would have resulted in a great loss of information about women in sexual unions. However, the definition tends to inflate some fertility estimates and depress contraceptive use indicators since women who have not been living with a man but are sexually active and successfully contracepting are not included. All women meeting these criteria who lived in a sampled household, including visitors, lodgers, domestic workers, and those temporarily absent were selected as eligible respondents 7. Including women on a de facto residency basis addressed the migration and work patterns of Black women, many of whom worked in nearby White urban areas for the day, or returned only on weekends or once a month. Others were involved in seasonal agricultural work, and during peak seasons temporarily lived on farms. By including all women in a household who met the qualifying criteria, whether the household was on a farm, in the White area, or a homeland, those who were missed at their permanent address had some likelihood of being sampled at their temporary place of employment or residence (Mostert, 1994a). The strategy also provided a chance of selection into the sample for women who were visiting or for some other reason were not at their usual place of residence. To provide independent demographic results for each survey area, sample sizes had to be sufficiently large to capture enough rare events (e.g., births and deaths) for robust analyses. Table 2 provides a list of the survey areas and the specified sample sizes. For each of the fourteen survey area, target sample sizes did not fall below 1000 respondents. 2. Sample Design Each sample was a stratified, random probability sample designed to be self-weighting and was drawn by the Centre for Statistics under the direction of Professor DJ Stoker at the HSRC. The sample frame used for the survey was the 1985 South African Census and the 1985 censuses for the TBVC states 8. While the official published figures for the censuses were not adjusted, researchers at the HSRC estimated adjustments before drawing the sample. What follows is a general description of the sampling procedure used for the each of the surveys in the SADHS. Since different information was available for each race group (e.g., addresses were not available for most Blacks), the sampling strategy differed by racial group. Even across Black survey areas, however, the sampling strategies varied to accommodate local conditions and availability of information. In each survey area the primary level of stratification distinguished between urban and rural areas 9. Another level of strata was sometimes constructed for a survey area, for example by district, depending upon availability of 6 In Lebowa, a slightly different definition of the target population was employed. See details on Lebowa in the appendix. 7 Interviewers for the non-black samples were instructed not to include Black domestic workers on the household coversheet. Information on these workers was collected on a separate form. 8 The 1985 South African census was taken March 6, 1985, on a de facto basis, and coverage included the selfgoverning homelands. The TBVC states undertook their own surveys simultaneously with some degree of financial and technical aid from the government of the Republic of South Africa (van Zyl, 1988:18). An analysis of the 1985 census undertaken by the HSRC showed considerable undercounts: 8.1% of Indians, 3.6% of Coloureds, 5.6% of Whites and 19.5% of Blacks were under-enumerated (Mostert and Hofmeyer, 1987). Sadie (1988) provides underenumeration estimates to the 1985 census that are similar but slightly lower. 9 The definition of urban and rural is ambiguous in South Africa, especially in Black areas. Many Blacks live in concentrated settlements and are not dependent on agricultural production for their livelihoods. Yet because there is no local authority (an administrative designation), these settlements are officially classified as rural. Functional urbanization, especially in the homelands, is considerably higher than the level indicated by official statistics.

9 5 information, population size and geographical layout of a given area. The sampling proceeded in two stages. At the first stage, enumeration sub-districts (ESDs) 10, small geographic units used in the census (between households), were selected; the second stage involved the selection of households within sampled ESDs. Each ESD was previously identified as urban or rural on the data tapes used in the sampling process. For each (sub)stratum, ESDs were listed and randomly selected in proportion to population size. The sampling fraction was based on the target sample size divided by the population size (females, age 12-49) in the stratum. Smaller ESDs were aggregated into one unit to increase the probability of selection. If the unit was drawn, one of the ESDs was selected as representative of the others. In other cases, for example KwaZulu, extra ESDs were generated and held in reserve, to be used in the event that some areas became inaccessible due to weather or unrest (Owen, 1994). Comparing the data tapes with the fieldwork schedules, it appears that ESD substitution occurred twice in KwaZulu, an area that experienced considerable unrest and flooding during the survey. Once the number of women to be interviewed from each sampled ESD was determined, that number was divided by the estimated number of women per household to obtain the number of households to contact. Since little information existed on the woman-household ratios, estimates based on previous research and guestimates, especially for the homelands, served as a guide. For example, the target sample size for KwaZulu was 1500 women. In urban areas (about a third of the population), the number of eligible women per household was estimated to be 2.5, in rural areas, 4.0. The number of households for the entire sample could be estimated by: 500 urban women x household 2.5 eligible women rural women x household 4.0 eligible women 450 households The number of households to be contacted in an ESD was calculated and then rounded to a multiple of the interviewing team size, usually eight in number. For example, if the proportionate allocation of households yielded an ESD with 15 sampled households, this would be rounded to 16 households and each interviewer in a team of eight would then be assigned two households. Rounding in this way, however, does not strictly adhere to principles of self-weighting sampling. The decision to use this technique was taken because of limited resources and relative inaccessibility of many areas (and the associated costs in locating disparate households in remote areas). Since there is no reason to suspect that rounding occurred in a biased manner, it is unlikely to have substantially altered the representativeness of the sample. Maps of the selected ESDs were available from the Central Statistical Service (the organization responsible for the census) and were reproduced for the fieldwork staff. These maps served to identify the geographical boundaries of the selected ESDs. The households were systematically selected in each ESD based on block listings compiled by the fieldwork staff or from a list of addresses available from the census (more details on the selection of households are provided in the following section). Households were then assigned to interviewers who interviewed all eligible women residing there. 3. Sample Implementation and Selection of Households by Survey Area The preceding section provides general outlines on the sample design for the survey areas of the SADHS. Most survey areas, however, had particular features that required some tailoring of these general sampling procedures, especially with respect to the identification of households. The following provides information on each of these areas to the extent that information about them is available. This section discusses the four population groups surveyed in the White RSA in urban and agricultural settings and then provides a general overview for the homeland areas. (See the appendix for details about each homeland survey.) RSA Samples: White, Coloured, and Indian There is a surprisingly little documentation on the non-african samples, or just about any other dimension of the survey pertaining to these population groups. In part, this may be due to the decentralized organization of fieldwork 10 For the 1991 census ESD s were renamed enumerator areas (EA s).

10 6 for the non-black samples (regional offices were delegated a great deal of responsibility for the data collection for these samples) which probably contributed to the lack of proper record-keeping and archiving (Mostert, 1994a). There is some information available through internal consistency checks and the coversheet data set (see below, response rates and substitution ) and the following description, based largely on interviews with Professor Mostert and Professor Stoker, provides a few more details. There is, unfortunately, almost no documentation to support the description. Sampling for the non-black population groups was based on the 1985 census and the addresses used for the census. Samples were drawn on the basis of race in the White RSA. Mostert (1994) points out that still at the end of the 1980s, though pass laws were dropped in 1986, there was very little overlap of race by residence and drawing racebased ESDs was not a problem. The next strata was the development region 11, and within development region urban and rural ESDs were selected proportional to population size 12. As described above, all eligible women in a household were interviewed, and the number of households to be selected was based on an estimated womanhousehold ratio. The sample of ESDs for all non-black groups was drawn by the national office in Pretoria by the HSRC. This office also was responsible for the fieldwork for the White population (though sometimes in cooperation with the regional offices). The selection of households and the fieldwork for the Indians and Coloureds, however, was delegated to the regional offices in which a majority of the respective groups resided. The HSRC office in Durban was responsible for the fieldwork for the Indian population, and for Coloureds, the office in Cape Town completed the work. The main office in Pretoria supplied the regional offices with the sampled ESDs and the regional offices, following instructions from Pretoria, drew up the address list and selected the households 13. RSA Samples: Black Urban Areas Most ESD maps for Black urban areas demarcated stands' or the plots of land in townships allocated for presumably one household. Based on the number of stands (and an estimated number of eligible women per household) an appropriate sampling interval was chosen and stands were systematically selected for the sample. The selected households were then assigned to interviewers. Where the stand demarcations deviated substantially from those listed on the map or where no detailed maps were available, fieldwork organizers blocklisted the ESD. This approach for selecting ESDs and then identifying households within them in Black urban areas had at least two major shortcomings: new squatter areas did not have a chance to be selected into the sample if they did not exist for the 1985 census, and any growth since 1985 in existing urban settlements was largely ignored. That is, some of the growth in Black urban since the 1985 census was likely to have taken the form of increased density per 'stand'; assuming one household per stand was not correct in many cases in Black areas. Though the previous government tried to control the number of Africans living in urban areas, the severe housing shortage produced many instances of multiple family housing units and overcrowding within households. In the case of new squatter areas, locating new settlements and assigning them a probability of selection required greater resources than were available for the study. The research team decided not to update the sample frame to include this new growth. Squatter settlements that occurred within a selected ESD, however, were incorporated into the sample. A field organizer estimated the number of households within the squatter areas, though blocklisting was often difficult due to suspicious residents, volatile conditions and the irregular structure and layout of these areas. In consultation with the Centre for Statistics, households in the new settlement were selected into the sample. While 11 There were nine development regions, geographic units developed for economic planning purposes by the Development Bank of Southern Africa. At that time, there was no administrative or political authority attached to these units, though the boundaries are quite close to the post-apartheid provincial designations. 12 There were some exceptions, however. If in a given ESD, the population of the race of interest was too small, it was not used because of the expense of finding so few people. 13 The census did not include address information for Coloureds living in rural areas on White farms, primarily in the areas North and East of Cape Town. The selection of this group was very similar to Black agricultural workers in the 'White' RSA (please refer to the description below).

11 7 there is no explicit code in the data specifying squatter areas, analysis of the completed interviews shows that about six percent of the Black-RSA sample resided in shacks (dwelling-type) in urban areas 14. This figure is certainly a rough estimate, and though it is probably an underestimate, it does indicate that some squatter areas were included. In the case where multiple households are found on one stand, blocklisting procedures may have helped somewhat. Those ESDs that were blocklisted by field organizers may have included these other units. However, at the time of the survey the researchers did not realize the extent of multiple household stands and did not provide any explicit guidelines for the fieldworkers to identify or select specific households under these circumstances. The effect of this sampling strategy on the representativeness of the sample for these urban areas depends on the characteristics of persons dwelling there information that did not exist at that time, and even now is quite limited. However, several reasonable assumptions can be made: 1. Residents of squatter areas or of second or third households on a stand are likely to be either newly arrived rural migrants or long-term urban dwellers who are unable to afford better arrangements. 2. In the case of the multiple-household stands, it may also be that the residents are family members and differ little from others in the community. 3. Squatter settlements have poorer sanitation and health services than townships. 4. Squatter settlements may not have better health and sanitary services than rural areas. 5. Residents of squatter areas are likely to be underrepresented in the SADHS. 6. Persons residing in second or third dwellings on a stand are also likely to be underrepresented. Overall, the data from the Black urban samples probably underestimate fertility and mortality levels and inflate contraceptive use rates. The Black urban sample is most likely disproportionately comprised of longer-term and more established township dwellers rather than squatters or non-primary households in township areas. Urban Blacks are more likely to have better sanitation, better access to health and family planning services, and are more likely to have and to desire fewer children than squatters or residents of second or third dwellings on stands. However, the dearth of data on characteristics of these populations makes a firm evaluation of the SADHS sample virtually impossible. We do not know to what extent the SADHS differs from the true population profile of Black urban dwellers. Indeed, selecting a representative sample of urban Black groups in formerly White South Africa has been quite difficult for all studies and an issue of some debate 15. RSA Samples: Black and Coloured Agricultural Workers Many Blacks and Coloureds were hired as both seasonal and permanent workers on farms in the White RSA. The population distribution of this group was known for each ESD, and the ESDs were then drawn to be proportionate to population size. The number of eligible women working on each farm in the ESD was estimated and farms were then systematically selected. No a priori definition of what constituted a household on White farms was established since type, structure and size of living quarters for temporary or seasonal workers varied tremendously. Interviewers were instructed to choose women from each farm in as random a fashion as possible, however, the extent to which this random selection occurred is unknown. Farmers were not always cooperative, and sometimes 14 Township houses were built by the government. Most of these houses were detached, single-family type structures constructed out of brick, complete with small front and backyards. These type of houses were presumably not coded as 'shacks.' 15 For an example of the difficulties of estimating the population size of a Black urban area, see Chris Steel (Pty) Ltd, et al, 1993.

12 8 identified the women to be interviewed for the interviewer (van Zyl, 1994b). Unfortunately, there is no way to determine the extent of this practice. It is likely that if this did happen, women would be more likely to be nonworking, due to pregnancy, disability, or age. RSA Samples: Domestic Workers An attempt was made to include a sample of approximately 125 domestic workers in the survey. However, no sample frame existed with which to randomly select respondents, and therefore, selection was essentially done on a quota basis. Homelands: Black Urban Areas Addresses were selected by first blocklisting all households within a selected ESD. Under the direction of a supervisor from the Opinion Survey Centre(OSC), the unit charged with the SADHS fieldwork, and in consultation from the Centre for Statistics at the HSRC, a systematic sample of households was selected from the list. Homelands: Black Rural Areas No address listings were available for these samples. For each ESD selected in the sampling process, field organizers drove throughout the area to estimate the number of households. A systematic sample of households was generated using a random starting point. Questionnaires Two models of questionnaires were used in the SADHS, the coversheet questionnaire and the individual questionnaire. The coversheet questionnaire was used to determine the eligibility of women in a sampled household, while the individual questionnaire was based on USAID/Macro DHS questionnaires. The household questionnaires collected information from a responsible adult in a sampled household on all women from ages who usually stay in the household, whether or not they are temporarily absent. Age, number of children ever born, the number of children born during the previous five years, the date of birth of the last child born, pregnancy status of each woman and the total number of persons living in the household were collected. Interviewers were then instructed to identify those women between the ages of 12 and 49 inclusive who were or are living with a partner, or who have had children, or who are currently pregnant as qualifying women for the individual questionnaire. Unfortunately, union status was not collected on the coversheet questionnaire which meant that there is no way of determining directly the number of eligible women as listed on the coversheet. Interviewers determined marital status through conversation and circled the corresponding line numbers for all women who qualified. The coversheets collected the same information for Black and non-black samples for all women age in the household; however, coversheets for Black households contained different response codes than those for non-black groups. The individual questionnaires, adapted versions of the USAID/Macro DHS surveys, collected information on the following topics: Respondent s background Water and toilet facilities Reproductive history Contraceptive use and sterilization Breastfeeding practices, including abstinence during breastfeeding Child health and immunization Experience of living with parents and in-laws Fertility preferences Husband/partner education and occupation The fourteen questionnaires are not identical across surveys, however. Most differences occur between the Black and non-black questionnaires. Specifically, Blacks did not receive questions on dissatisfaction with contraceptives,

13 9 husband s/partner s opinion on the matter, or what method of contraception women would prefer to use in the future. The questions on sterilization also differ considerably, most likely because sterilization rates among the Black population remain quite low, so little space was dedicated to this issue for African survey areas. Some questions have different response categories. The question regarding current use of contraceptives is a notable and important example. All questionnaires have the following question: Are you currently doing something or using any method to avoid getting pregnant? The questionnaires for the Black samples contain these response categories: 1. Yes 2. No, not using. Non-Black questionnaires presented a different set of options: 1. Not sexually active 2. Yes. 3. No, not using. Differences in questionnaires across racial groups are most common, although differences do occur across Black questionnaires as well. The questionnaire used in Lebowa, for example, has no category for never been married on the marital status question as the others do 16. Other inconsistencies in skip patterns and response categories also exist, and researchers should be attentive to variations when using the data. While this variation is unfortunate, some of these differences are the result of experiences in the field and attempts to smooth these problems in subsequent surveys (i.e., in those of the fourteen surveys not yet conducted). While the interviewers were trained using translated versions of the questionnaire, the interviewing itself used questionnaires printed in English or Afrikaans. Interviewers were asked to translate the questions in the local language or dialect when interviewing respondents. (See sections on Fieldworker Recruitment and Training and Problems and Complications in the Field below.) Pretesting A pretest of the SADHS was conducted from March, 1987, in non-sampled areas of Pretoria. Revisions and changes to the questionnaire were made under the direction of Dr. Mostert and Dr. Roussouw of the HSRC (Mostert, 1994). Input for these revisions also came from discussions held by experienced fieldwork staff (Greeff, 1987a). Fieldworker Recruitment and Training Fieldwork organization differed for Black and non-black groups and is discussed separately. For the Black samples, prospective fieldworkers with a minimum Standard 10 certificate (completed secondary education) were recruited by radio advertisement or through a local organization. Applicants were then required to pass an aptitude test to be selected and trained as fieldworkers. Two or three reserves per fieldwork team were also selected at that time and completed the training course. In general, because of the difficult economic situation and high unemployment, applicants held high qualifications (Greeff, 1994a). For the non-black samples, fieldworkers with prior experience in the Opinion Survey Centre (OSC), the HSRC fieldwork unit, were called upon to conduct the interviews in their areas. All interviewers were female (most fieldwork supervisors were male), and the interviews were matched-race and in most cases, matched-language. 16 Comparing distributions of the marital status question across survey areas indicates that interviewers probably coded 'never been married' women in Lebowa as 'no longer living together.'

14 10 The training course usually lasted three days and consisted of one day for general information and two days dedicated to the questionnaire. The last two days involved going through the questions one by one, going over translations and particularly important questions, how to do consistency checks, and any other issues which arose through discussion. Interviewers were asked to practice completing the questionnaire on a family member or friend before the fieldwork began. The training was conducted by senior OSC staff and generally a researcher from the demographic unit of HSRC was present to provide input and answer questions. Interviewers were instructed to visit households three times if qualifying women in the household were absent. They were further instructed not to substitute addresses in the event that no one was home at the designated household (Mostert, 1994a). The interviewer handbook for the non-black fieldwork indicates that additional addresses were supplied to interviewers to be used only in the event of vacancy or no contact after three attempts (Opinion Survey Centre, 1987). Interviewers were paid for each woman questionnaire and each child chart that was completed in an effort to minimize missing data information. This payment method, however, also encourages substitution. The problem of substitution is discussed in more detail below. Because of the multiplicity of languages, it was decided that it was neither cost effective nor organizationally feasible to translate questionnaires for all languages. For the Black areas, the questionnaires were printed in English and interviewers were asked to translate each question for the respondent. In training, a translated questionnaire was used to specifically address language difficulties that might arise in the field. Many of the Black interviewers were fluent in more than one language. If interviewers could not speak the language of a particular respondent, they reported the case to the supervisor who re-assigned another interviewer with the appropriate language skills. For the non-black samples, questionnaires were in both English and Afrikaans and administered in the language preferred by the respondent. Fieldwork As mentioned above, the Opinion Survey Centre (OSC), the fieldwork unit of HSRC, was contracted to conduct the interviews for the SADHS. The OSC was responsible for recruiting, hiring, and training interviewers, blocklistng, household selection when appropriate (and in consultation with the Centre for Statistics), transport of interviewers, monitoring progress in the field, quality control, and related fieldwork activities. 1. Black Fieldwork Field work was organized by teams, usually four or eight interviewers each, for data collection in Black areas. Each team was headed by a team organizer, a member of the OSC-HSRC who was responsible for transporting fieldworkers to their areas and for writing regular field reports on the performance of the members of the team, problems they might have encountered, and the realization of the sample assigned to that team. As discussed above in the sampling section, these organizers were also responsible for visiting each sampled enumerator sub-district or block and listing all households so that appropriate samples could be drawn based on the most recent information. One of these organizers, usually the most senior of the organizers, also acted as the head field organizer charged with coordinating field activity. The field supervisor, a senior staff member of the HSRC unit, was charged with drawing up a fieldwork schedule based on the sample specifications drawn by the Centre for Statistics at HSRC and systematizing the households to be contacted based on the household listings provided from the organizers. The supervisor was also responsible for gaining permission to conduct the survey from the local authorities and chiefs, monitoring training and fieldwork in general, and consulting with the principle investigator at the HSRC. The questionnaires were continually checked and edited throughout the fieldwork period by either a senior fieldworker or an appointed controller. If problems arose, they were either corrected at that time or, if necessary, the fieldworker was sent back to collect the appropriate information. If an interviewer had a poor performance, one of

15 11 the stand-by interviewers was asked to take her place. The questionnaires were then sent to either a regional office or the head office where a sample would again be checked and verified. 2. Non-Black Fieldwork The interviewing for the White, Coloured, and Indian samples was carried out on a regional basis, with each regional office (Cape Town, Durban, and Pretoria) responsible for selection and training of interviewers as well as data collection and control. As mentioned previously, interviewers for these surveys were generally seasoned interviewers with prior experience with the Opinion Survey Centre. The interviewers from each region were given a list of household addresses to interview within about a one-month period. No interview teams were used. Opinion Survey Centre personnel contacted interviewers daily or every other day to monitor progress. After completing a certain portion of interviews, the interviewers were instructed to send completed questionnaires to the regional office for control and checking (van Zyl, 1994b; OSC, no date). Fieldwork Schedule Table 3 provides the schedule of field dates for the surveys and the expected and final sample sizes for each survey area. The fourteen surveys were conducted over the course of two years to accommodate the schedule of the fieldwork organization. Table 3 shows that the target and realized sample sizes are fairly close for most surveys. In several instances, however, there are notable differences. In the case of Lebowa, this is due to the initial definition of eligibility in combination with under-estimated woman-household ratios. In Gazankulu, the woman-household estimates were the source of the larger than expected sample size. No record exists for Bophuthatswana, but a problem with the ratios is likely to have been the case there as well. For the non-black samples, there is no record on why the realized samples fell short of the targets. There is evidence which suggests, however, that the deficit is the portion of sampled households without completed interviews. (See response rate below.) Problems and Complications in Fieldwork The HSRC conducted the fieldwork in accordance with the procedures outlined above. However, the social and political circumstances of apartheid operating in the late 1980s made fieldwork conditions less than ideal. Furthermore, many researchers in South Africa commonly used substitution in their fieldwork, and substitution may have also occurred during data collection for the SADHS, especially given the difficult field conditions. This section evaluates the evidence available for each of these issues, beginning with substitution. 1. Substitution The problem of substitution, interviewing women at non-sampled households (for example, if no one was home, eligible, or cooperative in sampled households) is a problem that plagues research in South Africa. Substitution apparently was a common practice and indeed a recent survey supported by the Kaiser foundation makes this an explicit part of its methodology (Herschowitz and Orkin, 1995:7) Nonetheless, substitution can seriously bias the results of a survey since women who are at home and are willing to be interviewed are most likely systematically different from those who are not home or who are reluctant to provide an interview. This section looks explicitly at the evidence that might indicate the presence of substitution in fieldwork and to what extent it may have occurred. The training provided to all fieldwork staff and interviewers specified that no substitution of the selected households

16 12 should be done (Mostert, 1994a) 17. Professor Mostert notes that once he indicated there should be no substitution, the fieldwork organization doubled its price quotation. Nonetheless, some substitution appears to have occurred. Substitution information was recorded on the household coversheets, but was data-entered only for the non-black samples 18. The coversheets have since been destroyed and it is now impossible to estimate the extent of substitution for the Black samples. Table 4 provides a summary of the substitute address information for White, Coloured and Indian samples. In each case, less than 6% of households were substituted representing less than 6% of the total completed interviews for each group. In the Black-RSA sample, less than 2% of the coversheets contained any address information at all and only one coversheet was coded as a substitute address. This may reflect a decision not to code the information rather than an absence of substitution altogether. No information exists for the other survey areas, that is, for the homelands. Ultimately, we know very little about the extent of substitution for the Black samples, except to note, again, that the principle investigator insisted that this not be allowed. Discussions with persons involved with fieldwork indicate that little substitution occurred, especially in most homeland areas since usually someone in the household was home, and cooperation was generally high (Greeff, 1994a; Mostert, 1994a; van Zyl, 1994b). This is further supported by scattered information in field reports; many indicate that no substitution occurred. However, the field reports do not exist for every survey area, and those that do exist do not always mention substitution at all. To the extent substitution occurred among the Black samples, it was probably most prevalent in urban areas where unrest is greater and finding women at home and finding cooperative respondents is more difficult. The tendency to substitute under these conditions, however, may have been countered by the tighter field supervision in urban areas. That is, work in urban areas was easier for field organizers to monitor because of density and infrastructure. If substitution did occur, it would likely depress contraceptive use rates and inflate fertility rates since women at home would be more likely to be unemployed or pregnant. The effect on child and infant health and mortality is less clear. Children whose mothers are at home may be better cared for and more closely supervised. It may also mean that there is other regular income (e.g., from a partner or spouse) that supports the household and consequently a relatively higher standard of living. It could also indicate a household subject to a high degree of male authority and little autonomy for the woman. While wealthier households are usually healthier, lower female autonomy is associated with poor health outcomes for children in many settings. Conversely, women who are not at home are presumably working. A woman who has secured employment is likely to be more educated or at least more likely to be knowledgeable about and to gain access to appropriate medical facilities than women who are at home. Education, employment and knowledge and familiarity with health facilities are all positively associated with improved health outcomes for children in other contexts. The implications of substitution are serious since the sample is no longer representative. Though fieldworkers were given express instructions not to substitute, there is evidence that this happened anyway. For non-black samples, assuming the coversheet data are correct, this does not appear to be more than 6% of all completed interviews. For the Black samples, we know relatively little about the extent of substitution, though field reports and persons involved in the study indicate that substitution rarely if ever took place. Substitution is very difficult to verify from the data themselves. For example, it is likely that working women would be underrepresented, but there is no reasonable external comparison. Further, few studies conducted elsewhere document difficulties with substitution; standards do not exist by which to judge these data. In the author s view, substitution may have occurred, but the admittedly scant evidence suggests that this did not happen to a damaging degree. Discarding the data on the basis 17 A training manual for a study conducted by the demographic research unit for the Population Development Programme specified that if no one is home after the second visit, the household to the left can be substituted, and all substitutions should be appropriately coded (HSRC, 1991). While a training manual from another study does not provide conclusive evidence about substitution in the SADHS, it does indicate that substitution was a part of methodology in some studies of the HSRC. 18 Professor Mostert has indicated that the word 'substitution' was left on the coversheet from previous work done by OSC. He further notes that 'substitution' for the SADHS really means 'additional' - pre-selected addresses or households that were used in the event of vacancy or no contact after three attempts (Mostert 1994a).

17 13 of rumored substitution seems unwarranted. However, analysts using these data should recognize that biases may have been introduced by substitution. 2. Political Unrest, Weather, and Other Adverse Conditions Encountered in the Field Problems in the field specific to a given area usually revolved around issues of weather and road conditions, distances, work schedules of potential respondents, sensitivity of questions, and cooperation with the local authorities. The information presented in this section has been taken from field reports, usually written by field team leaders in charge of a small group of interviewers. These reports are not ideal sources: they were written for supervisors, so there may have been an incentive not to elaborate on problems or difficulties; they provide little detail on the extent of the problems or how the issues were resolved; they are not available for every survey; and the subject matter of the reports varies considerably across survey areas and interview teams. These documents nonetheless provide some indication about the conditions in the field from the point of view of field team leaders who worked closely with the interviewers and who experienced the local conditions in the field. Field conditions in the four TBVC states are considered first, then the six SGT homelands, and finally the field conditions of the surveys for the four racial groups residing in the RSA. There was only one fieldwork report available for Transkei which only mentioned poorly maintained roads as a hindrance to fieldwork. Indeed, the field team organizer reported that respondents were very receptive (Ngceba, 1987). In general the survey seemed to go well in Venda, though there were some difficulties related to the content of the survey. Venda is a fairly traditional culture; questions regarding fertility can be offensive (special note was made that post-partum sexual taboos and contraceptive methods questions were particularly difficult for respondents to answer 19 ), and it is taboo to discuss child sickness and death (Motlhamme, 1989; Radebe, 1989; Sibanyoni, 1989). A related noteworthy comment comes from Dr. J. McCutcheon, Director-General of the Department of Health for Venda (1989) who notes that question about the timing of the previous child 20 will probably be interpreted to mean Did the husband want the pregnancy.. as many women do not consider themselves to have any choice in such matters. An HSRC OSC representative responding to his letter indicated that this would be taken up in training (Richards, 1989). Another team leader mentioned that rumors about muti killings (deaths perceived to occur at the hands of a witch who uses parts of the body for medicines) fueled suspicions for both interviewers and respondents (Matlala, 1989). Mr. Matlala also reported that women in many areas were either hoeing or working at factories which made interviewing difficult, especially in remote areas. Women interviewers in Venda did not wish to work past dark which, after travel time, left few hours for locating working women and then interviewing them. He reports that there was only one substitute used by his team. All team leaders report that distances were sometimes long with bad road conditions and the villages at times hard to find. In Ciskei, Rambau (1988), and Jaca (1988) both note that the police would not sign the authority forms because they contained the emblem of the South African police 21, but that otherwise they did not run into problems. One anonymous field report from the Ciskei (no date) mentions that respondent cooperation varied considerably from area to area, but that in all eventually the work was done. This report also mentioned that Ciskei has many converted villages and that much of the rural population has been resettled there, an observation referring to the former government s program to resettle Blacks in Black areas and 19 Q323: A custom exists among some couples not to resume sexual relations while the mother is still breastfeeding. Did you practice this custom after the birth of (NAME)? Q326: To achieve a desired time/space between births or to avoid having too many children, couples use various ways and methods to avoid pregnancy. Have you ever used any of these methods? (List of methods follows). 20 Q325: Did you want your last (current) pregnancy then (now), did you want it at a later stage, or did you want it sooner or not at all? 21 Standard procedure included notifying the local authorities prior to beginning survey activities. Local authorities included police, sometimes political parties, and in many areas, chiefs or headmen. The incident described here is indicative of the animosity felt among the various homeland and RSA police and security forces.

18 14 to develop rural areas in homelands. There are no field reports on Bophuthatswana. In Gazankulu, a self-governing territory, the police were also not cooperative. This area contained a great number of Mozambican refugees 22, and some respondents gave their address as Venda instead of Gazankulu, but otherwise the fieldwork went smoothly (Greeff, 1987c). Aerial photography was used in KaNgwane and created confusion for one team (Morifi, 1989), but improved village and household identification greatly for another (Mathebula, 1989). Since this was new technology at the time, such conflicting reports are understandable. Mr. Mathebula also comments that field control for his team was very difficult because of the inaccessibility of the roads (especially Eerstehoek and Nkomazi). Again in KwaNdebele, fieldworkers had a great deal of (unspecified) trouble with the police (Mongolobotho, 1988); otherwise no problems are mentioned in the reports. The survey in KwaZulu was plagued by heavy rain falls and bad road conditions (Greeff, 1987d). In his methodological description, Mostert (no date) notes that fieldwork in KwaZulu was hampered by political unrest and assassinations ; however, Greeff (1994b), the field project organizer, states that the unrest was not so severe but related to faction fighting and homeland politics (p.2). In Lebowa, the field report is concerned primarily with the woman-household ratio (see above in samplin ), but also mentions poor roads and language problems. Additionally, fieldworkers had difficulty contacting women near Pietersburg since many worked days, and near Sekgosese, Bolobedu, Moderong and Mapulaneng, many of the women worked on farms returning only on weekends or less often (Greeff, 1987e). Finally, in QwaQwa, there seemed to be problems with the quality of the interviewers' work, but controls were in place to check on this regularly. (Matlala, 1988; Maboea, 1988; Greeff, 1988; Sibanyoni, 1988). There are almost no field reports available for the surveys undertaken in the RSA. There are several reports from field leaders for the Coloured sample in the Cape area which indicate that things ran smoothly (Valley, 1988). There are no materials available for the White or Indian fieldwork. The evidence available from these fieldwork reports is by no means comprehensive and are certainly not conclusive about the affect field conditions might have had on data quality. However, they do not provide any indication that the obstacles encountered in the field were handled in such a way that might be injurious to data quality or sample design. Indeed, statements about difficulties finding working women and quality control, for example, suggest that instructions regarding revisits to households to find eligible women were being followed and that supervisory structures in the field were in place. Response Rates Tables 5 and 6 provide estimated response rates by survey area for the SADHS for Black and non-black samples, respectively. These numbers are calculated using the coversheet data set and as noted above, two different versions of the coversheet were used for Black and non-black households. Turning to Table 5, the table for the Black samples, the response rates are generally high (79% to 93%), but they are comparable to response rates found in the USAID/Macro DHS surveys conducted in other African countries (and in fact, are generally a bit lower). The small numbers refusing interviews on the basis of subject matter or objections to the HSRC further corroborate the field report notes indicating general cooperation of respondents in homeland areas. As noted on the table, no information on household-level refusals was collected on the coversheet for the Black samples. Table 6 presents analogous information for the non-black samples. The woman-level response rate, based on coversheets with valid response code information, is extremely high (more than 96%), and for all three groups, nonresponse counts are strikingly low. However, the non-black coversheet dataset included coversheets that contained no valid response codes (and hence not included in the woman-level counts) but nonetheless were assigned a record number. These coversheets are marked as missing on the table and constitute a sizable proportion of all non-black coversheets, especially for Whites. It is unclear what these represent. One explanation is that there were no qualifying women found at these households, though this seems unlikely to be the case for so many households and 22 No indication is given whether the refugees were interviewed or not. Though they probably spoke Portuguese, a contraindication to being interviewed, they may have also spoken Shangaan, a language indigenous to that area.

19 15 in any case, the appropriate response code should have been entered. It is possible that no one was at home after the three revisits required for a selected household, but again, an appropriate response code should have been recorded (unfortunately, the number of visits a household received was not collected on the coversheets). The coversheets might also be associated with sampled ESDs that were for some reason not used in the data collection process. However, an analysis of the geographic information shows that the missing coversheets are not concentrated in any one area, but in almost all cases they are associated with ESDs for which there are also completed interviews. Another possibility is that these coversheets simply might have been unused and somehow mistakenly included with completed coversheets to be data-entered. However, a number of these coversheets with invalid response codes do contain some data on women within the household. Table 7 provides a summary of selected variables from these missing coversheets. For the Whites, about 20% of the coversheets with invalid response codes have some information, for the other two groups, roughly a third contain some information. This suggests another explanation: at least some of these coversheets represent sampled non-cooperating or non-responsive households. Presuming the worst case, that al missing coversheets represent non-cooperating households, response rates were re-estimated on the basis of the achieved woman-household ratio and the number of missing coversheets. The adjusted 54% response rate for the White population seriously compromises this sample (although this rate is not unlike those of urban areas in the United States). For the Indian and Coloured groups, the adjusted response rate is not nearly so damaging; rates of 89% and 85%, respectively, are still reasonable 23. Again, the adjusted response rate provides a worst-case scenario since some unknown number of these coversheets likely represent households which should be validly excluded from the sample. Interviewing Hours Table 8 summarizes the times of the day when completed interviews were begun according to coversheet information. The bulk of the interviewing occurred in mid to late afternoon. Since work days often begin and end early in South Africa, this indicates that women who work during the day may have been likely to be home. The exceptions are Venda and Gazankulu. In Venda, the field reports indicate that interviewers did not work late hours. The field reports do not mention anything like this for Gazankulu, but this is also a relatively traditional area. Gazankulu was also likely to be an area in which local female interviewers were not able to work extended hours. Data Entry and Editing An outside organization was employed to enter the survey data 24. The data were subject to a series of programs designed to edit and clean and were then reviewed for final editing by HSRC researchers (van Zyl, 1994b). Under the direction of Dr. L. van Tonder, at that time a researcher in the demographic unit at the HSRC, the fourteen surveys were combined and a record layout of the data file was produced. Considering the complexity of the aggregation involved, the resultant data file reflects a great deal of care and attention to detail. For example, the 23 The severity of this problem for Whites relative to the other two groups is notable. Part of the explanation may be again the decentralized nature of the fieldwork for these population groups, though one might presume that since the head office was largely responsible for the White sample, the quality control would have been more rigorous. If the 'missing' coversheets are indeed refusals, part of the explanation might have been the deep concern for security on the part of Whites during the late 1980s. However, it is likely that this concern was shared by other groups as well and so is unlikely to explain the large discrepancy. 24 It is unknown if the data were keyed in twice, an expensive but effective way to catch data entry mistakes. Errors were likely even with very conscientious data entering, but in general, they are extremely difficult to find. Some data entry errors in the SADHS were readily detectable, for example, no household identification numbers were entered for the non-black sample and all partners/spouses in KaNgwane were entered as business men (a highly unlikely possibility for any homeland). Data entry mistakes also appeared in the complex geographical hierarchies used to indicate the location of each respondent.

20 16 response categories for the question on whether the respondent had received a post-school qualification were yes and no, but in some surveys, yes was coded one, while in others it was coded two. A simple frequency of this variable by survey area shows that the coding differences were made consistent when the files were combined. Indeed, the differences in the response categories for the current contraceptive use question for Blacks and non-blacks (see above in 'Questionnaire') was not changed in the data tapes, but the coding scheme is clearly documented on the record layout. After combining the files, researchers at the HSRC constructed a series of recodes based on the data and attached these to the data file. These recodes are presumably analogous to the standard recode files produced by Macro International for the USAID-DHS surveys and were most likely used for reports and later publications. Unfortunately, while the program constructing the variables is available, documentation explaining definitions, the cases included in the denominator, or coding schemes does not exist. Weights Each of the 14 surveys was designed to be self-weighting, the sample design did not include over-sampling or strategic sampling of certain groups. Within each survey area, that is, each homeland and for the four population groups in the White RSA, the sample was designed to be representative and no weights are needed for independent analyses of any of the 14 surveys. Used as a pooled dataset, however, the fourteen samples do not represent the national distribution of the population without the application of proper weights. No documentation exists regarding the calculation of these weights, but the values placed on the data records correspond to the estimated distribution of the population in South Africa. Table 9 presents the weighted and unweighted distribution of the SADHS by survey area and compares them with census figures for the years 1991 and Overall, the percentage allocation of the weighted sample closely approximates the estimates of the 1991 figures which are adjusted. The close match provides further evidence that HSRC researchers adjusted the 1985 census figures before drawing the samples. Data Quality The quality of the data is not influenced solely by sampling and data collection and entry procedures. Inaccurate answers given by respondents or responses incorrectly coded by interviewers are two sources other than methodology which might introduce bias into survey data. The following section presents results from data quality assessments using the individual and household (coversheet) datasets. The methodological report on the assessment of data quality for the USAID/Macro DHS surveys (IRD, 1990) serves as a guide to the analyses undertaken in this section. Several measures IRD employed in their comparative assessment of the Macro DHS data are applied to the SADHS data, although, due to data restrictions, sometimes a modified version is used. The extent to which the SADHS data were cleaned or missing data imputed is unknown. Some of the measures used in this section are sensitive to the degree of imputation; interpretation of quality assessments based on these measures may thus vary depending on the assumptions regarding imputation. For most analyses, each of the fourteen samples is treated independently because they were collected separately and therefore might be subject to a different set of influences. In most cases, aggregated measures also are presented (i.e., total, Black, and non-black samples) and these estimates are based on weighted data. The assessments presented here are not intended to be exhaustive, but rather to provide an initial investigation into the extent that data quality issues found in other studies also may be present in the SADHS. In keeping with the focus of the thesis, emphasis is placed on the selection of eligible women into the sample and on the quality of fertility data (i.e., birth histories and related questions). Data quality assessments of health and mortality information also are important, but these topics are beyond the scope of this paper. 1. Coversheet Information and Respondent Selection Analysis of the coversheet dataset often provides useful information about the household composition of the interviewed women. Coversheet information was collected from a knowledgeable adult in the household and wrong information gathered at this point might have resulted in the inappropriate selection or exclusion of respondents. For

21 17 example, if informants underestimated young women s ages, eligible women age 12 may have been excluded. The coversheet questionnaires did not collect information on all household members so that complete analyses of systematic downward estimates of young ages are not possible. However, interviewers were instructed to collect information on women age in the household (for the coversheet), and to interview those eligible age (using the individual questionnaire). If interviewers were aging women so they would have to interview fewer women in the household, heaping at the age of 50 might occur. In fact, the converse was true with these data. While numbers were small in each survey area at the older ages, a tendency to heap on 49 occurred in all samples except in KaNgwane, Bophuthatswana, and the Ciskei (results not shown). This is likely to be a result of the payment scheme. Since interviewers were paid per interview, it would have been to their advantage to interview as many women in a household as possible. Making older women younger, then, would have increased both the number of eligible women and interviewers subsequent remuneration. Since older women were at the end of their reproductive years, the effect of this bias would tend to lower contraceptive use rates and increase some fertility estimates. However, as noted, the number of women at these higher ages was small and any effect is likely to be minimal 25. Information on the coversheet and in the questionnaires also allows for an analysis of de facto coverage of women. Interviewers may have tended to exclude women who were not considered to be regular members of the household (for example, visitors or temporary lodgers). The number of women absent was coded on the coversheet in the form of response codes. The individual questionnaire asked women if their usual household was elsewhere. Comparing the aggregate totals of these two numbers provides a rough indication of the de facto coverage of eligible women in a household. In other words, the totals for women who were absent from the household (as reported on the coversheet, refer to Tables 5 and 6 26 ) should be greater than (since some were refusals or non-response) but correspond roughly to the number of women interviewed but who stated their usual residence was elsewhere. Aggregating across all subsamples (since women could theoretically have been visiting or working anywhere in South Africa), the coversheets indicated that 2731 women were reported absent, and 2117 women were interviewed who were not staying at their usual residences. The difference yields a balance of about 600 women reported as absent and not picked up as visitors at other households. While this is a crude estimation of absent and visiting eligible women, the 2117 women who were not at their usual place of residence (about 10 percent of the total sample) indicate that women were interviewed on a de facto basis and that this number was not wholly incongruous with the total reported as absent. 2. Age Heaping and Distribution of Samples by Age Categories Once women have been selected into the sample, the interviewing process may produce errors in coding either due to incomplete or inaccurate information given by the respondent, or the miscoding of information on the part of the interviewer. One common problem in many developing country settings is that of age heaping; respondents may tend to report their ages as numbers ending in zero or five. Table 10 provides Whipple s index of digit preference for age reporting by survey area. Those values in excess of one indicate a preference for ages ending in five or zero. The values range between.97 (Bophuthatswana) and 1.26 (Gazankulu). With the exception of Gazankulu and the Ciskei (1.24), all other indices are 1.16 or below indicating that age heaping was not a substantial problem for most survey areas. The panels on Chart 1 present the distribution of women interviewed by single years of age for each survey area as an indication of the extent of age heaping in the SADHS. These charts show that when age heaping did occur, it did not always correspond to numbers ending in five or zero. In KwaZulu (panel C), for example, ages 20, 25, and 30 showed characteristic signs of heaping, but so did ages 27 and 32. In KwaNdebele 25 Age, marital status, and reproductive experience were used in combination to determine the eligibility of women. Since the coversheet did not include information on marital status, however, it is not possible to extract the necessary eligibility information to compare with completed interviews. 26 Since the differences among the response codes 'R Not available,' 'Visiting elsewhere' or 'Unavailable due to employment' may have been difficult for interviewers to distinguish given the information they received, they have been combined into one category for this comparison.

22 18 (panel D), heaping occurred at the ages of 21 and 29. Women in KaNgwane (panel F), tended to favor age 30, but also age 24. Though age heaping does not appear to be a serious problem in these data, the charts do illustrate that the Black samples were made up of a greater proportion of women at younger ages (<30) relative to those at older ages. A gradual decline in the proportions of women at older ages in the sample is expected because of mortality and slightly larger cohorts of younger women. However, an unusually disproportionate distribution of ages indicates the possible exclusion of a select group of women, for example, women who were working. Table 11 shows the distribution of the sample by age groups. As a preliminary comment, less than 25 cases of young women aged appeared in the total dataset even though fieldworkers were instructed to interview eligible women in this age category 27. The low numbers may indicate that very few women in this age were in fact eligible, or that the interviewers chose not to or were not allowed to interview these girls. Note, however, that a code for parental refusal for the Black samples did not exist (Table 5), and very few such codes were recorded for the non-black samples on Table 5 6. Since this age group is commonly excluded from most demographic analyses, the low numbers do not necessarily introduce bias. Ages also represented a fairly small portion of the distribution, though again, this was not unexpected since many this age would not have met the definition used in the selection of eligible women. This category was especially small for the non-black samples and may reflect in part the stigma attached to teenage pregnancies in these groups, and the consequent reluctance to admit to pre-marital births. In general in the Black samples, younger women age represented a substantial portion of the sample in each area. In KaNgwane, for example, year olds constituted almost 30 percent of the sample. This is likely to reflect in part the fact that this age group might have been easier to find since most were no longer in school or many may have been home with childrearing responsibilities. It may also be in part due to age misreporting (for example, a tendency to associate age 20 with the age at first birth). The distributions, however, were quite similar to those found in the USAID/Macro DHS surveys conducted in other sub-saharan African countries (IRD,1990:28). 3. Quality of Birth History Data A major objective of the SADHS was to provide information on fertility. The construction of fertility estimates often relies on data collected in the fertility histories of each woman who reported that she had ever given birth. Each woman was asked to provide basic information on every child she had borne, starting with her first born. Completeness and accuracy of the birth histories are required for reliable fertility estimates. As noted above, neither the available study documentation nor the data have an indication of the type or extent of data cleaning or imputing. The results of the assessments undertaken here, then, may produce two different sets of interpretations, depending on whether imputation or cleaning occurred. Below, the completeness, the consistency, and heaping tendencies in the data are examined. Each assessment is discussed individually and a more general discussion of quality concludes the section.. The data from South Africa show that very little missing information exists in birth histories, and the information recorded there for the most part was internally consistent. Women were asked to report the numbers of children living at home, dead, and living away from the respondent. They were then asked about each child individually, providing the interviewer with information on that child's sex, age, birthdate, survival status, where the child lives, and age of death of the child when appropriate. The totals women reported in the first section of questions corresponds almost exactly to the number of children listed in the fertility history (and indeed, interviewers were instructed to make sure the numbers agree). For each child listed in the birth histories, very few cases with missing birth month, year, or age exist in the data the information was complete for at least 98 percent of children for each survey area. The completeness of the information may indicate that a part of the cleaning and editing process included imputation or correcting for cases with missing or inconsistent information. On the other hand, figures from other USAID/Macro DHS studies conducted in southern Africa showed similarly high figures for completed (nonimputed) information (IRD, 1990:57-8,85). Comparing the reported ages of children with the ages calculated on the 27 Eligible women were defined as those aged who were living with or had ever lived with a man, or were currently pregnant or had ever given birth.

23 19 basis of their birthdates and the interview date also shows a high correspondence. More than 90 percent of reported ages for children were consistent with the reported birthdate at every parity, and this figure increased if age and birthdate differences within one year were included (results not shown). Displacement of births may also affect the estimates of fertility indicators. The placement of the first birth in this regard is particularly important since subsequent births are often placed in reference to this date. Table 12 presents indices of birth heaping on years ending in zero or five, a common problem found in other third world settings. The index represents the cumulative portion of births in a year ending in zero or five (in this case 1960, 1965, 1970, 1975 and 1980) relative to the proportion in two years prior through two years after the year in question. The year 1985 was not used to calculate the index since some surveys were conducted in 1987 and births occurring in 1985 would have been censored. The first year of first-birth heaping considered here, 1980, was at least seven years prior to the earliest interview and may have been as many as nine years earlier for some survey areas. The indices, then, are likely to be subject to recall biases. A value of one on the index indicates no heaping; a value higher than one means an excess of births was reported for those years; a value less than one indicates a deficit of births reported. As a benchmark, the IRD methodological report stated that values higher than 1.05 indicate significant heaping (1990:60). Table 12 shows that with only five exceptions, first-birth heaping was a problem in all survey areas by the IRD criterion, whereas only a few countries in the IRD report showed heaping to this degree. Heaping was more prominent for first births that occurred in earlier years (at greater durations since first birth), but heaping of earlier births cannot completely explain the high scores of the indices. The high correspondence between age and birthdate together with the propensity to report first births on particular years suggests that interviewers may have corrected (or filled in) ages based on birthdate. This may also indicate that the data were edited such that ages were recalculated based on birthdate; however, not all ages and birthdates correspond perfectly and it is unlikely that an editing program would correct some ages and not others. Since Table 12 examined only first births, an assessment of all births and their placements might provide information on overall patterns in birth histories. For example, heaping may occur for first child born, but once this date is fixed, subsequent births may be measured correctly thereafter. Table 13 provides an abbreviated assessment (combined samples only) on the tendency to heap children s reported ages on years ending in zero or five. These age ratios calculate the number of children at a given age relative to the average reported at the age one year above and one year below that age. The ratios show that respondents did not tend to prefer ages ending in zero or five when reporting their children s ages, except for the age of 15. Ages one, three, six, eight and twelve seemed to have been favored in these data. The relatively high index values for age one probably indicate a combination of low reporting of children born within the last twelve months and a tendency to report them as age one. The heaping on age six is likely to be a function of the questionnaire design and is addressed below. Heaping of other ages is more difficult to explain, but it is probably more likely due to preferred birth years than to favored ages. So, for example, for those surveys conducted in parts of 1987 and 1988, age twelve coincided with a birth year of Indices of heaping on birth years ending in zero or five for all births are presented in Table 14. The indices show that some groups, for example those in Gazankulu and KwaZulu, exhibited a preference for years ending in zero. For KaNgwane and the Transkei, heaping was highest for the most recent year considered, 1980, a result not easily explained. However, the aggregate index of heaping is lower for most survey areas than its counterpart for first births. This suggests that respondents may have preferred years ending in zero or five to fix the first birth date, but may have reported subsequent births in intervals from that date, that is, with a decreased propensity to favor specific years. Respondents may also have a tendency to report births according to number of years before the interview, for example, a respondent may indicate that a child was born five years ago. Chart 2 shows this relationship for each survey area (panels A-E), and panel F shows weighted aggregate patterns. The charts demonstrate that in general no distinguishable patterns exist, except some areas have a notable increase in the number of births six years prior to the survey relative to five years prior to the survey. This may be explained in part by the design of the DHS-type surveys which asked a module of questions on child health and illness for those children aged five or less. To avoid a long and tedious battery of questions, interviewers may have aged children such that they were older than five. The pattern may also, in part, explain why 1980 was a favored year for reporting births, since interviews conducted in 1987 might have aged five year olds to a birthdate in Table 15 calculates birth year ratios for five and six

24 20 calendar years prior to the interview date. These indices, analogous to the age ratios discussed above, show the number of births in the fifth (or sixth) year before the interview relative to the average number of births the year preceding and the year following the fifth (or sixth) year. Gazankulu, Bophuthatswana, and the White samples all show tendencies to heap on the sixth year, but otherwise the indices for the fifth year have a value of.90 or greater. 4. Discussion of Data Quality In sum, the quality of the SADHS data was measured along several different dimensions. Information from the coversheet questionnaires proved to be limiting since the instrument did not contain a complete household roster or marital status of women in the household. From the response code information, however, the number of absent women exceeded the number of women interviewed at households other than their own (e.g. visitors), but not by an unusually high amount. Age heaping, another common problem in surveys conducted in developing countries, also did not appear to be a substantial problem in most survey areas in these data. The analysis of birth information collected in birth histories showed that the histories contain almost complete information on birthdates and ages for children. Furthermore, the vast majority of reported ages correctly reflected the stated birthdate. Other demographic studies in southern Africa also showed a high level of complete information on birth histories, and the high level of completion in South Africa was not, therefore, particularly unusual. However, the reader should be aware that the type or extent of data cleaning or imputation for these data is unknown. The high degree of consistency in the South African data may have been due to mechanical editing or imputation procedures, or it may have been due to interviewers who completed the child s history on the basis of partial information given to her by the respondent. However, imputation procedures themselves do not necessarily introduce bias nor do they necessarily correct biases that might appear in the data beforehand. That is, even if ages were imputed based on birthdates, if birthdates were heaped in a particular year, the imputed ages also would have shown this tendency. The investigation into birth histories suggested that women in these samples tended to favor first-birth birth years ending in zero or five. They favored these years for other births as well, but not to the same degree. This suggests that these respondents, after reporting a first-birth birthdate, recalled subsequent births in terms of intervals. Indeed, the distribution of reported children s ages showed that they did not tend to heap on ages ending in five or zero, but instead the ages coincided with a preference for certain years. In general, the Black samples tended to heap more than non-black samples, though considerable variation in distribution patterns by survey area existed. Finally, interviewers might have reported five year olds as six to avoid child health questions in some areas, but this aging did not appear to be extensive. Aging children was a greater problem in the non-black samples than in Black samples, especially for Whites. The impact the displacement of births can have on fertility estimates depends on the type and extent of displacement as well as the indicator under consideration. For example, estimates of recent fertility are based on births in the preceding 60 months, that is, those children age 0 to 4. If children who are five are recorded as six, this will have little effect on recent fertility. However, biases are introduced if displacement occurs such that children s births are transferred across age boundaries. In these data, Lebowa and Ciskei showed signs of heaping on age 5 (see Chart 2, panels A and C). If this was due to four-year olds being reported as five, recent fertility in these areas may have been underestimated. Arnold (IRD, 1990:93) points out that displacement may also exert an impact on estimates of fertility change over time. Shifting births back could have the effect of underestimating recent fertility and overestimating fertility for the prior period, say 5-9 years before the survey, thus simulating an apparent decline in fertility. The data of the SADHS appear to be of reasonable quality. Though they are by no means ideal, in most respects they produce distributions and tendencies which resemble those in the USAID/Macro DHS samples for southern Africa. The assessments discussed here, again, are not exhaustive, but they provide some indication of the types and magnitudes of possible biases that appear across survey areas. These data can contribute important

25 21 information on fertility and reproductive patterns in South Africa, but their use requires a great deal of care 28 Conclusions The SADHS conducted by the HSRC provides important information regarding women s reproductive history, child health and mortality, and the general socio-economic circumstances of women during a time for which few other demographic data sources are available. The reluctance of the HSRC to make the data available for outside quality assessment, and the lack of methodological documentation have, however, produced skepticism regarding the reliability of the data. And indeed, the study was conducted under less than ideal circumstances: organizations (such as the HSRC) associated with the government evoked a high degree of suspicion and sometimes fear, especially in Black urban populations; researchers were subject to international isolation and received no technical or financial support; and many areas experienced high levels of political unrest. While these factors produced a hostile climate in which to implement research, the study suffers from some neglected or financially unfeasible methodological aspects independent of the political context. Identical questionnaires across geographical areas and population groups, questionnaires translated into the major language groups, better designed coversheets, and careful documentation on sample design, training, fieldwork, and quality control checks would have substantially the understanding of the strengths and weaknesses of these data. The issues raised in this paper show that in fact, to the extent possible, sample design and interviewing techniques accommodated the difficult conditions of sampling and fieldwork and for the most part do not introduce substantial bias into the representativeness of the sample. This is not to suggest there are no biases. Black urban populations are probably not properly represented, for example. However, information from field reports, sampling notes, and from the data tapes themselves indicate that the study was not subject to gross violations of recognized survey methodological principles. The critique of the methodology of the South African Demographic and Health Survey may be summarized as follows: 1. Little documentation was completed on the White, Coloured, and Indian samples. Each of these samples should be used with extreme care and subjected to thorough data quality assessments. The White sample, for example, may have a response rate as low as 54 percent; however, the distributions of various characteristics of each group were consistent with expectations. 2. The Black urban sample probably reflects a wealthier and more established population than was true because main households in townships were probably interviewed disproportionately more than secondary households or squatters. In general, Black urban areas and especially squatter settlements are most likely underrepresented due to sample frame limitations. 3. Some substitution probably occurred in all samples, although fieldworkers were given explicit instructions not to do so. For non-black samples, the evidence indicates that substituted households represent less than six percent of completed interviews. There is no information on the extent to which substitution occurred in Black samples. The evidence suggests that the practice was minimal in homeland and rural areas It was probably more likely to occur in Black urban areas in the White RSA since women in these areas would be more likely to be away working, and uncooperative respondents are more common. 4. Considering the size and scope of the dataset, the data entry process and combining of the surveys produced reasonably clean data. However, differences in skip patterns, response categories, and general questionnaire design vary across survey areas and greatly complicate analysis. 28 As noted, the health and mortality sections of this dataset probably also contain useful information, but data quality assessments of this information is beyond the scope of this project.

26 22 5. The assessment of the quality of data showed that completeness of information was generally high, though completeness may be due to editing or imputing. The assessment also showed that some digit preference for women s ages and on birth histories occurred, primarily in the form of heaping on birth years ending in zero or five. Most data quality checks showed patterns similar to those found for other southern African countries participating in the USAID/Macro DHS surveys (IRD, 1990). 6. In spite of methodological shortcomings and hazardous fieldwork conditions, careful analysis and presentation of results based on these data can provide useful and important information regarding the demographic processes of South Africans in the late 1980s. The intention of this paper was to present the methodology of the SADHS in honest and forthright terms. The problems arising from the design and implementation of the study documented here, while not insubstantial, do not make these data unusable. Indeed, the SADHS contains important information regarding demographic processes in the late apartheid era. The evaluation was hampered by an inability to compare the strengths and weaknesses of the SADHS to other studies beyond basic data quality measures; few studies systematically document difficulties in fieldwork, or adaptive strategies used to overcome them. Ultimately, no exacting measure of study quality can be assigned to the SADHS. Nonetheless, the value of the study will increase as more detailed and varied data quality assessments are published and the strengths and weaknesses of the data are addressed and accommodated in analyses. Responsible use of these data will provide important insights into the history of fertility processes, health conditions, and mortality in South Africa, information vital to planning and providing for adequate services in a transitional society.

27 23 References Anonymous Fieldwork Weekly Report. Team fieldwork report for the SADHS-I, Ciskei (handwritten). HSRC, South Africa. Anonymous. no date. Middledrift. Partial team fieldwork report for the SADHS-I, Ciskei (handwritten). HSRC, South Africa. Caldwell, John and Pat Caldwell The South African Fertility Decline. Population and Development Review. 19(2) Calitz, J.M. and M.J. Grove A Regional Profile of the Southern African Population and Its urban and Non- Urban Distribution, Halfway House: Development Bank of Southern Africa. Chris Steel Architest (Pty) Ltd, the Centre for Health Policy, and Rosmarin and Associates A Guide Plan for Primary Health Care Services In Greater Soweto. Phase II, Volume 3: Annexures. Johannesburg: Centre for health Policy[?]. Greeff, DP. 1987a. "Fertility Survey BNBC09T034: Training/Discussion Session on Questionnaire." Internal memo from MarkData, the fieldwork unit of HSRC, Pretoria, March 17, Greeff, DP. 1987b. "Verslag: Loodsstudie vir Fertiliteitsopname BNBC09T034." Internal memo from Opinion Survey Centre, the fieldwork unit of HSRC, Pretoria, March 25, Greeff, DP. 1987c. Veldwerkverslag. Official fieldwork report for the SADHS-I, Gazankulu, compiled by the field project manager. HSRC, South Africa. Greeff, DP. 1987d. Veldwerkverslag, Demografiese en Gesondheidsopname. Official fieldwork report for the SADHS-I, KwaZulu, compiled by the field project manager. HSRC, South Africa. Greeff, DP. 1987e. Veldwerkverslag. Official fieldwork report for the SADHS-I, Lebowa, compiled by the field project manager. HSRC, South Africa. Greeff, DP Algemene Kontrole Verslag. General control report for the SADHS-I, QwaQwa, compiled by the field project manager. HSRC, South Africa. Greeff, DP. 1994a. Interview with Carol Kaufman, 10 February, University of Pretoria, Pretoria, South Africa. Greeff, DP. 1994b. Untitled. Comments and observations as chief field organizer of the SADHS-I on the methodology report by WP Mostert, DHS1: Methodology (no date). HSRC, Pretoria, South Africa. Human Sciences Research Council. No date. HSRC-OSC Manual for Interviewers. Document of the Opinion Research Center of the HSRC. Institute for Resource Development An Assessment of DHS-I Data Quality. DHS Methodological Reports 1 and 2. Columbia, Maryland: Institute for Resource Development/Macro Systems, Inc. Jaca, V.E Week 1 & 2 Field Report. Team fieldreport of the SDHS-I, Ciskei (handwritten). HSRC, South Africa. Kuzwayo, Ellen Call Me Woman. Johannesburg: Ravan Press.

28 24 Maboea. E.T Teams 4: Control Report. Team fieldwork report for the DADHS-I, QwaQwa. HSRC, South Africa. Marks, Shula and Stanley Trapido (eds.) The Politics of Race, Class and Nationalism in Twentieth-Century South Africa. Cape Town: Oxford University Press. Mathebula, B.I Population Survey in KaNgwane. Team fieldwork report for the SADHS-I, KaNgwane (handwritten). HSRC, South Africa. Matlala, V Fieldwork Report. Team fieldwork report for the SADHS-I, Venda (handwritten). HSRC, South Africa. McCutcheon, J Letter to Mrs. Richards of Opinion Survey Centre of HSRC. 23 February, Mongolobotho, M.M Demographic and Health Survey. Team fieldwork report for the SADHS-I, KwaNdebele (handwritten). HSRC, South Africa. Morifi, S.S KaNgwane Population Survey Fieldwork Report. Team fieldwork report for the SADHS-I, Venda (handwritten). HSRC, South Africa. Mostert, WP and BE Hofmeyer "Socioeconomic Factors Affecting Fertility in the Developing Countries and of the Developing Population Groups in South Africa." In Southern African Journal of Demography. 2(1):1-6. Mostert, WP. 1994a. Head of Demographic Research Unit, Human Sciences Research Council, Interview conducted with Carol Kaufman. February 24, Human Sciences Research Council, Pretoria. Mostert, WP. 1994b. Methodology DHSI. Notes. June, HSRC, Pretoria, South Africa. Mostert, WP. No date. "DHS1: Methodology." Chapter in an unpublished report on the South African Demographic and Health Survey. Pretoria: HSRC. Motlhamme, S Fieldwork Report. Team fieldwork report for the SAHS-I, Venda (handwritten). HSRC, South Africa. Ngceba, Victor C.Z Demographic and Health Survey, Report on the Survey. Team field report, Transkei (handwritten). HSRC, South Africa. Odendaal, E. 1988? Projekverslag Kleuringe. Project report on the Coloured population of the RSA. HSRC, South Africa. Opinion Survey Center "Manual for Interviewers. Project BNC09T035/36: Demographic and Health Survey. Human Sciences Research Council. Owen, Rena Formerly a statistician at the Centre for Statistics at the HSRC. Interview conducted with author. University of Pretoria, Pretoria. Platzky, Laurine and Cherryl Walker The Surplus People: Forced Removals in South Africa. Johannesburg: Ravan Press. Platzky, Laurine and Cherryl Walker The Surplus People: Forced Removals in South Africa. Johannesburg: Ravan Press. Radebe Fieldwork Report. Team fieldwork report, SADHS-I, Venda (handwritten). HSRC, South Africa.

29 25 Rambau Field Report. Team field report of the SADHS-I, Ciskei. HSRC, South Africa. Republic of South Africa, Department of Health. Annual Reports over years Pretoria: Government Printers. Richards, T Letter to Dr. J. McCutcheon, Director-General of the Department of Health, Republic of Venda. 1 March, Sadie, JL A Reconstruction and Projection of Demographic Movements in the RSA and TBVC Countries. Report no Pretoria: Bureau of Market Research, University of South Africa. Sadie, JL "The Consequences of Rapid Population Growth." In WP Mostert and JM Lotter (eds) South Africa's Demographic Future. Pretoria: HSRC. Sibanyoni, P.A Control Report. Team fieldwork report, SADHS-I, QwaQwa. HSRC, South Africa. Sibanyoni, P.A Fieldwork Report. Team fieldwork report, SADHS-I, Venda (handwritten). HSRC, South Africa. Stoker, D.J Steekproefontwerp van die Fertiliteitsteekproef in die Republiek van Transkei. Formal description of the sample design for the Transkei. HSRC, South Africa. Stoker, D.J. 1988/9. Venda Sample. Notes on Venda sample design, SADHS-I, Pretoria, South Africa. Stoker, D.J Former Director of Centre for Statistics at the HSRC. Interview conducted with author. September 1, University of Pretoria, Pretoria. Valley, O Projeckverslag / Project Report. A complilation of field reports by many anonymous field leaders for the SADHS-I, Coloured-RSA (handwritten forms). HSRC, South Africa. van Zyl, JA Adjustment of the 1985 Census Population of the RSA by District. Research report no Pretoria: Bureau of Market Research, University of South Africa. van Zyl, JA. No date. "1. Introduction" and "2. The Sample." Brief reports on the methodology of the SADHS. No publisher information. van Zyl, Johan History, Scope and Methodology of Fertility and Family Planning Surveys in South Africa. Paper presented at the Annual meetings of the Population Association of America, Miami, May, van Zyl, JA. 1994b, Research Scientist, Demographic Unit, HSRC. Series of discussions with author. Human Sciences Research Council, Pretoria. Wilson, Francis and Mamphela Ramphele Uprooting Poverty: The South African Challenge. Cape Town: David Philip. Worden, Nigel The Making of Modern South Africa: Conquest, Segregation and Apartheid. Oxford: Blackwell.

30 26 Appendix Description of Sample Design by Individual Homeland Areas Transkei The sample for the Transkei was formally described by Professor DJ Stoker. The sample incorporated a greater degree of detail than that for most other survey areas. For example, the Transkei sample design used estimated woman-household ratios unique to each district as opposed to average urban and rural estimates used in most other areas. Bophuthatswana No information is available on the sampling methods used for Bophuthatswana. The final sample size is 10% larger than expected and may indicate that the estimated woman-household ratio was too small. This may also suggest that interviewers were following instructions, that is, they were in fact interviewing all eligible women in the household. Venda No notable differences in sampling methodology are known, but the settlement patterns facilitated the listing of households in an ESD. The people of Venda live in small gatherings throughout a hilly area. Field organizers were able to draw maps of each of these settlements and then systematically select households within each settlement. Ciskei Rather than use a simple urban/rural dichotomy as the first level of stratification, four degrees of urbanization were distinguished: metropolitan, regional centers, sub-regional centers, and rural areas. The same method of sampling was used as described in chapter five. Lebowa In Lebowa, the first SADHS survey to be conducted, a broader criterion of marriage was used at first: Any woman within the ages of 12 to 49 who was currently or had ever been in a steady relationship (i.e., sexually active) was considered eligible. The result was a considerably higher number of eligible women per household than anticipated. After four days in the field and in consultation with the Centre for Statistics, unfinished sampled ESDs were completed and a new field program was constructed. In an attempt to contain costs and not compromise the sample, the remaining ESDs were scaled back to half their size, that is, four households were interviewed in each sampled ESD instead of eight as originally planned. The principle investigators at this point decided to change the condition for eligibility to living with a partner/husband instead of just sexually active. There were two main reasons for this decision: First, there was some sensitivity over the identification of sexually active women within the household; second, many of the women who qualified under this criterion were young, not in a steady sexual relationship, and had no children. While this is important information, samples with these respondent characteristics included little or no experience with fertility, contraception, or infant and child mortality. To satisfy the objectives of the survey, that is, to provide a sample with sufficient numbers for analyses on rare events, a new definition of eligibility was adopted (described in the text in the 'target population' section) (Mostert, 1994b). Gazankulu No information on sampling was available for Gazankulu. Like Lebowa, the first few field days yielded a higher than expected woman-household ratio. Again, in consultation with the Centre for Statistics, the sample was scaled back to contain costs. KwaZulu No explicit explanation of sampling method existed. Notes on KwaZulu indicated that 18 strata were used to allocate households proportionally to the population. A comparison of the ESDs on the fieldwork program with those on the data tapes indicated that two ESDs were substituted, most likely due to weather or political unrest.

31 27 KwaNdebele No explicit notes on the sampling were available. QwaQwa QwaQwa is a densely populated, small and highly homogenous former homeland. The first level of stratification for QwaQwa was trichotomous: Urban, semi-urban and rural. Households were allocated proportionately to the population for each of the three strata. It was decided beforehand that sampled ESDs should contain three clusters of eight households each except for the urban area (Phuthaditjaba) where twelve city blocks were sampled to include four households each. KaNgwane The sampling procedure was the same as outlined above, but additionally, the seven magisterial districts were used as substrata within urban/rural divisions. Aerial photographs were used to locate settlements and identify households.

32 Table 1 Estimated Population Distribution of South Africa, by Geo-political Unit, 1991 % of Total South Total African Population Sex Ratio White RSA* White Black Coloured Indian RSA Total Self-governing Territories* Gazankulu KaNgwane KwaNdebele KwaZulu Lebowa QwaQwa SGT Total Independent States** Transkei Bophuthatswana Venda Ciskei TBVC Total South Africa Total * From the 1991 Republic of South Africa Population Census. Report no (1991). Pretoria: Government Printers. ** From SATBVC Statistical Abstracts, Halfway House: Development Bank of Southern Africa. Population figures are based on 1989 estimates.

33 Table 2 Target Sample Sizes of the Component Surveys of the SADHS.. Target Sample Size Self-Governing Territories: Lebowa 1500 Gazankulu 1000 KwaZulu 1500 KwaNdebele 1000 KaNgwane 1000 QwaQwa 1000 Independent States: Transkei 1500 Bophuthatswana 1500 Venda 1000 Ciskei 1000 RSA: White 2000 Indian 2000 Coloured 2000 Black 4000 Total:

34 Table 3. Survey Date, Target Sample Size and Completed Number of Interviews by Survey Area, SADHS.. Target Completed N Interviews Self-Governing Territories: 1-26 June, 1987 Lebowa July, 1987 Gazankulu September, 1987 KwaZulu May, 1988 KwaNdebele Jan - 10 Feb, 1989 KaNgwane July, 1988 QwaQwa Independent States: 7-25 March, 1988 Ciskei September, 1988 Bophuthatswana Nov - 4 Dec, 1988 Transkei Mar - 14 Apr, 1989 Venda RSA: 13 Aug - 5 Sept, 1987 White Aug - 9 Oct, 1987 Indian Oct - 28 Nov, 1987 Coloured February, 1988 Black Total:

35 Table 4. Substitution Address Information for non-black Population Groups Residing in the white RSA SADHS Whites Indians Coloureds Household Level # of HHs with Completed Interviews # with Substituted Addresses % of Households w/completed Interviews at Substituted Addresses Woman Level # of Completed Interviews* # of Completed Interviews at Substituted Addresses % of Completed Interviews at Substituted Addresses Source: Data taken from original coversheet datasets of the SADHS. This information was not available for groups living in the homelands and is virtually non-existent for the blacks within the RSA. * The differences between the completed number of interviews shown here and those for the individual level data tapes are most likely due to control and checking procedures, after-the-fact editing, or other data entry and editing processes.

36 Table SADHS Response Code Results, Black Samples Total RSA Independent Homelands Self-Governing Territories (Black) Black Bophutha- Ciskei Transkei Venda Gazankulu Kangwane Kwa- KwaZulu Lebowa QwaQwa tswana Ndebele Number of Households Individual Interviews: [1] Completed interviews * Woman-level Non-Response/Refusals Codes: R Not available Visiting elsewhere ~ Medical reasons Unavailable due to employment ** Refused screening information Objection to subject Objection to HSRC ~ Wrong respondent [2] Woman-level Non-response/Refusal Total Woman-level Response Rate (%) [1]/[1 + 2] Total N on Tapes Completed Interviews (as recorded on coversheet) - Tape N Source: SADHS coversheet dataset; SADHS woman-level dataset * The coversheet used for black samples did not collect data on household-level refusals/non-response ~ Not included in Non-response/refusal total. Excluded from woman-level response rate calculation ** Refused screening information could be an indication of a household-level refusal, but is treated here as an individual-level response cod

37 Table SADHS Response Code Results, Non-Black Samples Total RSA (non-black) White Coloured Indian Household Level Results [1] Total # of coversheets (1 cs/hh) [2] # of coversheets w/response code info missing coversheets % Households with valid coversheet info [2]/[1] Individual Interviews: [3] Completed interviews Woman-level Non-Response/Refusals Codes: R Not available Visiting elsewhere ~ Medical reasons Unavailable due to employment * Refused screening information Objection to subject Objection to HSRC ~ Wrong respondent [4] Woman-level Non-response/Refusal Total **Household-level response codes: Cannot communicate in Afrikaans/English Interview refused by parent No person qualifies Cannot make contact No one home for duration of survey Vacant household Used wrong respondent Household-level Non-response/Refusal Total *** Woman-level Response Rate (%) [3]/[3 + 4] **** Adjusted Woman-level Response Rate (%) Total N on Tapes Completed Interviews (as recorded on coversheet) minus Tape N Source: SADHS coversheet dataset; SADHS + Missing indicates that all response codes were either zero (invalid code) or missing, but the coversheet was assigned a record number suggesting an attempted contact of a household. ~ Not included in Non-response/refusal total. Excluded from woman-level response rate calculations. * Refused screening information could be an indication of a household-level refusal, but is treated here as an individual-level response code. ** Household level response codes are not included in woman-level response rate calculations. The information is provided here for completeness. *** Based on coversheets with complete response code information. **** Missing coversheets are assumed to be refusals and estimated woman-hh ratios are applied. Response rates are re-calculated based on new estimates of refusals. For example, for whites: (i) (1484 interviews + 8 refusals)/1456 valid coversheets ~ women per household (ii) women/hh * 1246 households ( missing coversheets) ~ 1275 women (iii) These 1275 women are considered refusals and entered into the response rate 1484 interviews /(1484 interviews + 8 refusals estimated refusals) ~ 53.6% (Equations reflect rounding)

38 Table 7 Distribution of Selected Variables of Missing Coversheets*, SADHS Whites-RSA Indian-RSA Colored-RSA Total # of missing coversheets # with some information Woman Characteristics: Age < Total Number of Women Who Gave Birth Number of Children Total Household Characteristics: Number in Household Number of Women Between in Household * Missing coversheets are those with missing or zero response codes but with record numbers recorded on them.

39 Table 8 Hours During Day Completed Interviews Were Conducted, SADHS* RSA (no homelands) Black White Coloured Indian N % N % N % N % Completed interviews Morning till 12: :00-13: :00-14: :00-15: :00-16: :00-17: :00-18: :00-19: :00-20: :00 and later Missing Independent States Transkei Bophuthatswana Venda Ciskei N % N % N % N % Completed interviews Morning till 12: :00-13: :00-14: :00-15: :00-16: :00-17: :00-18: :00-19: :00-20: Missing Self-Governing Homelands Gazankulu Kangwane KwaNdebele KwaZulu Lebowa QwaQwa N % N % N % N % N % N % Completed interviews #### Morning till 12: :00-13: :00-14: :00-15: :00-16: :00-17: :00-18: :00-19: :00-20: Missing Note: The codes available for interview times did not include the period 21:00 and later for the non-white population groups. * Data taken from the coversheet dataset of SADHS

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

41 Table 10 Whipple s Index of Digit Preference for Ages Ending in 0 or 5, SADHS Whipple s Index of Digit Preference* Total N ** Total ** Black samples ** Non-black samples Self-Governing Territories Lebowa Gazankulu KwaZulu KwaNdebele QwaQwa KaNgwane Independent States Transkei Bophuthatswana Venda Ciskei White RSA Black (non-homeland) Indian Coloured White X20 + X X40 + X45 * Whipple s Index of Digit Preference =. 2 *( X18 + X X46 + X47) where X i = the number of women reporting that they are age i. ** Based on weighted data

42 Table 11 Distribution of SADHS, by Age Categories (%) Age: Total * Total * Black Samples * Non-Black Samples Self-Governing Territories Lebowa Gazankulu KwaZulu KwaNdebele QwaQwa KaNgwane Independent States Transkei Bophuthatswana Venda Ciskei White RSA Black (no homelands) White Coloured Indian Note: In the total sample, there were less then 25 cases in categories and 50+. * Based on weighted data.

43 Chart 1 Age Distribution by Survey Area, SADHS [A] % of Females at Given Age Lebowa (n=1993) Age [B] Gazankulu 7 % of Females at Given Age (n=1175) Age [C] KwaZulu 7 % of Females at Given Age (n=1474) Age Source: SADHS (continued...)

44 Chart 1 Age Distribution by Survey Area, SADHS (continued) [D] % of Females at Given Age KwaNdebele (n=1060) Age [E] % of Females at Given Age QwaQwa (n=1004) Age [F] % of Females at Given Age KaNgwane (n=1002) Age Source: SADHS (continued...)

45 Chart 1 Age Distribution by Survey Area, SADHS (continued) [G] Transkei % of Females at Given Age (n=1583) Age [H] % of Females at Given Age Bophuthatswana (n=1646) Age [I] % of Females at Given Age Venda (n=1003) Age Source: SADHS (continued...)

46 Chart 1 Age Distribution by Survey Area, SADHS (continued) [J] Ciskei 15 % of Females at Given Age (n=1068) Age [K] % of Females at Given Age Blacks-RSA (n=3745) Age [L] % of Females at Given Age Coloured-RSA (n=1817) Age Source: SADHS (continued...)

47 Chart 1 Age Distribution by Survey Area, SADHS (continued) [M] Indians-RSA % of Females at Given Age (n=1788) Age [N] Whites-RSA % of Females at Given Age (n=1484) Age Source: SADHS

48 Table 12 Index of heaping on year of first birth ending in 0 or 5, SADHS Year of first birth index* # Women who had ever given birth ** Total ** Black samples ** Non-black samples Self-Governing Territories Lebowa Gazankulu KwaZulu KwaNdebele QwaQwa KaNgwane Independent States Transkei Bophuthatswana Venda Ciskei White RSA Black (non-homeland) Indian Coloured White xi * Index = i=1 (x i-2) + (x i-1) + x i + (x i+1) + (x i+2) where i 1 = 1960, i 2 = 1965, i 3 = 1970, i 4 = 1975, and i 5 = 1980, and x = number of women with first birth in year i. ** Based on weighted data.

49 Table 13 Age Ratios for Living Children by Single Year of Age (1-15), SADHS Total Black Samples Non-Black Samples Note: Based on weighted data.

50 Table 14 Index of Heaping of Reported Birth Year for Years Ending in 0 or 5, SADHS Index of Heaping for Reported Birth Year in**: Aggregate Index* *** Total *** Black samples *** Non-black samples Self-governing Territories Lebowa Gazankulu KwaZulu KwaNdebele QwaQwa KaNgwane Independent States Transkei Bophuthatswana Venda Ciskei White RSA Black (no homelands) Indian Coloured Whites * Aggregate Index = x i, where i 1 =1960, i 1 =1965, i = 1( xi 2) + ( xi 1) + xi + ( xi + 1) + ( xi + 2) i 1 =1970, i 1 =1975, i 1 =1980, and x= number of births reported in year I. ** Index of Year Heaping = x i, where x = number of births. 2 * [( xi 2) + ( xi 1) + xi + ( xi + 1) + ( xi + 2)] reported in year i. *** Based on weighted data.

51 Chart 2 Number of Births by Calendar Year Prior to Interview, SADHS [A] 400 Self-Governing Territories Number of Births Lebowa Gazankulu KwaZulu Years Prior to Interview [B] 400 Self-Governing Territories Number of Births KwaNdebele QwaQwa KaNgwane Years Prior to Interview (continued...)

52 Chart 2 Number of Births by Calendar Year Prior to Interview, SADHS (continued) [C] 400 Independent States Number of Births Transkei Bophuthatswana Venda Ciskei Years Prior to Interview [D] 400 White RSA (non-black Samples) Number of Births Coloured Indian White Years Prior to Interview (continued...)

53 Chart 2 Number of Births by Calendar Year Prior to Interview, SADHS (continued) [E] 700 White RSA 600 Number of Births RSA-Black Coloured Indian White Years Prior to Interview [F] 3500 Aggregate Distributions (Weighted) Number of Births Total Black Samples Non-Black Samples Years Prior to Interview

54 Table 15 Birth Year Ratios, SADHS Centered on Period*: 5 Years Before 6 Years Before the Interview the Interview ** Total ** Black Samples ** Non-Black Samples Self-Governing Territories Lebowa Gazankulu KwaZulu KwaNdebele QwaQwa KaNgwane Independent States Transkei Bophuthatswana Venda Ciskei White RSA Black (no homelands) Indian Coloured White Bx * Birth year ratio x years before the interview =,.5(B x-1 + B x+1) where B x = number of births in the xth year before the interview. ** Based on weighted data.

55