South African HIV/AIDS, Behavioural Risks, Sero-status, and Mass Media Impact Survey (SABSSM) 2002: Guardian data - All provinces

This project used the HSRC's master sample (HSRC 2002). A master sample is defined as a selection, for the purpose of repeated community or household surveys, of a probability sample of census enumeration areas throughout South Africa that are representative of the country's provincial, settlement and racial diversity. The sampling frame that was used in the design of the master sample was the 2001 census Enumerator Areas (EAs) from Statistics South Africa (Stats SA). The target population for this study were all people in South Africa, excluding persons in so-called special institutions (e.g. hospitals, military camps, old age homes, schools and university hostels). The EAs were used as the Primary Sampling Units (PSUs). Although the 2001 census data are not yet available, it was decided to use the 2001 EAs for the master sample because the sampling units would remain relevant for future surveys conducted by the HSRC within five to ten years' time. In addition, the HSRC would soon have access to the most recent census statistics over this period for weighting of future survey results, including this study. The sample was designed with two main explicit strata, namely, provinces and the geography type (geotype) of the EA. In the 2001 census, the four geotypes are urban formal, urban informal, rural formal (including commercial farms) and tribal areas (i.e. the deep rural areas). In the formal urban areas, race was also used as a third stratification variable. What this means is that the Master Sample has been designed to allow reporting of results (i.e. reporting domain) at a provincial, geotype and race level. A reporting domain is defined as that domain at which estimates of a population characteristic or variable should be of an acceptable precision for the presentation of survey results. The census 2001 EA data provided by Stats SA for drawing the sample contained an estimate of the number of dwelling units (DUs) or visiting points (VPs). A visiting point is defined as a separate (non-vacant) residential stand, address, structure, and flat in a block of flats or homestead. The 2001 estimate of visiting points was used as the Measure of Size (MOS) in the drawing of the sample. The visiting point is the Secondary Sampling Unit (SSU) in each of the selected PSUs. In this study, all people in all the households resident at the visiting point were initially listed, after which the eligible individual was randomly selected in each of the following three age groups 2-14, 15-24 and 25 years and older. These individuals constituted the Ultimate Sampling Units (USUs) of this study. Having completed the sample design, the sample was drawn with 1 000 PSUs or EAs being selected throughout South Africa (see Figure 2). These PSUs were allocated to each of the explicit strata. With a view to obtaining an approximately self-weighting sample of visiting points (i.e. SSUs), (a) the EAs were drawn with probability proportional to the size of the EA using the 2001 estimate of the number of visiting points in the EA database as a measure of size (MOS) and (b) to draw an equal number of visiting points (i.e. SSUs) from each drawn EA. An acceptable precision of estimates per reporting domain requires that a sample of sufficient size be drawn from each of the reporting domains. Consequently, a cluster of 11 VP was systematically selected on the aerial photography produced for each of the EAs in the master sample. Since it is not possible to determine on an aerial photograph whether a 'dwelling unit' is indeed a residential structure or whether it was occupied (i.e. people sleeping there), it was decided to form clusters of 11 dwelling units per PSU, allowing on average for one invalid dwelling unit in the cluster of 11 dwelling units. Previous experience at Statistics SA indicated a sample size of 10 households per PSU to be very efficient, balancing cost and efficiency. Overall, a total of 14 450 potential participants composed of 4 001 children, 3 720 youths and 6 729 adults were selected for the survey and 13 518 (93.6%) were actually visited. A small proportion (6.4%) of potential respondents could not be approached due to logistic constraints that were unavoidable in a study of such magnitude. Among the 13 518 individuals who were selected and contacted for the survey, 9 963 (73.7%) persons agreed to be interviewed, and 8 840 (65.4%) agreed to also give a specimen for an HIV test. The sample was designed with the view to enable reporting of the results on province level, on geography type area and on race of the respondent. The total sample size was limited by financial constraints, but based on other HSRC experience in sample surveys it was decided to aim at obtaining a minimum of 1 200 households per race group. In fact, the aim was to obtain 1 200 Indian households, 1 800 coloured households, 2 200 white households and 4 800 African households, a total thus of 10 000 households. The number of respondents per household for the study was expected to vary between one and three (one respondent in each of the three age groups). A 70% response rate was assumed and a HIV+ prevalence rate of 20%. However, the total refusal and noncontact rate was much higher than expected. Nevertheless, all cases where the interview could have been done were included in the analysis.

本项目采用了南非人类科学研究委员会（Human Sciences Research Council，HSRC）2002年主抽样样本。主抽样样本的定义为：为开展重复社区或家庭调查，从南非全境的普查区（Enumeration Areas，EAs）中抽取的概率抽样样本，该样本能够代表南非的省级、定居类型与种族多样性。本次主抽样样本设计所采用的抽样框，来自南非统计局（Statistics South Africa，Stats SA）发布的2001年普查区数据。本研究的目标总体为南非全境所有民众，但排除所谓“特殊机构”内的人员，例如医院、军营、养老院、中小学及大学宿舍内的人员。本次研究以普查区作为初级抽样单元（Primary Sampling Units，PSUs）。尽管2001年人口普查数据尚未完全公开，但研究团队仍决定采用2001年普查区构建主抽样样本，原因在于该抽样单元在未来5至10年内，仍可适用于HSRC开展的各类后续调查。此外，在该时间段内，HSRC将可获取最新的人口普查统计数据，用于包括本研究在内的后续调查结果的加权处理。本次抽样样本设计设置了两大显性分层变量，即省级行政区与普查区的地理类型（geotype）。2001年人口普查中，地理类型共分为四类：正规城市社区、非正规城市社区、正规农村社区（含商业农场）以及部落领地（即深度农村地区）。在正规城市社区中，种族被作为第三层分层变量。这意味着，本主抽样样本的设计可支持在省级、地理类型及种族层面发布调查结果（即报告域）。报告域的定义为：可将人口特征或变量的估计值控制在可接受精度范围内，用于展示调查结果的统计域。Stats SA提供的2001年普查区抽样数据中，包含了住宅单元（dwelling units，DUs）或到访点（visiting points，VPs）的数量估计值。到访点的定义为：独立存在的（非空置的）住宅地块、地址、建筑以及公寓楼或宅地内的公寓单元。本次抽样过程中，以2001年到访点数量估计值作为规模测度（Measure of Size，MOS）。在每个抽中的初级抽样单元中，到访点即为二级抽样单元（Secondary Sampling Unit，SSU）。本研究首先对到访点内所有住户的全部居民进行造册登记，随后从以下三个年龄组中分别随机抽取符合条件的个体：2~14岁儿童组、15~24岁青年组以及25岁及以上成年组。上述被抽取的个体即为本次研究的最终抽样单元（Ultimate Sampling Units，USUs）。完成抽样设计后，研究团队在南非全境抽取了1000个初级抽样单元（即普查区），详见图2。上述抽中的初级抽样单元被分配至各显性分层组中。为构建近似自加权的到访点（即二级抽样单元）样本，研究团队采取了两项抽样策略：（a）以2001年普查区数据库中的到访点数量估计值作为规模测度，按照与普查区规模成比例的概率抽取普查区；（b）从每个抽中的普查区内抽取等量的到访点（即二级抽样单元）。为保证各报告域内估计值达到可接受的精度，需从每个报告域中抽取足够规模的样本。因此，研究团队基于主抽样样本中每个普查区的航拍影像，系统抽取了由11个到访点组成的群集。由于无法通过航拍影像判断“住宅单元”是否为实际居住建筑，或是否有人员居住（即是否有人员在此就寝），研究团队决定每个初级抽样单元内设置11个住宅单元的群集，平均允许群集中存在1个无效住宅单元。南非统计局过往经验表明，每个初级抽样单元内抽取10户家庭的样本规模可在成本与效率间实现最优平衡。整体而言，本次调查共招募了14450名潜在受访者，其中包括4001名儿童、3720名青年与6729名成年人；实际完成走访的受访者共13518名，占比93.6%。由于此类大规模研究无法避免的后勤限制，仅有6.4%的潜在受访者未能完成走访对接。在完成对接的13518名受访者中，共有9963人（73.7%）同意接受访谈，其中8840人（65.4%）同意提供HIV检测样本。本次抽样样本的设计目标为支持按省级行政区、地理类型区域以及受访者种族三个维度发布调查结果。本次总样本规模受经费限制，但参考HSRC过往抽样调查经验，研究团队设定了各种族组至少抽取1200户家庭的目标。具体目标为：抽取1200户印度裔家庭、1800户混血家庭、2200户白人家庭以及4800户非洲裔家庭，总目标样本量共计10000户。本次研究预计每户家庭的受访者数量为1~3名（即三个年龄组各1名受访者）。研究最初假设的应答率为70%，HIV阳性患病率为20%，但实际拒绝率与未对接率均远高于预期。尽管如此，所有可完成访谈的案例均被纳入最终分析。