Sampling Procedure
The 1996 census sampling frame was used to select EAs for the 1997 Rural Survey but the survey was restricted to the former homeland areas, i.e. Transkei, Bophuthatswana, Venda, and Ciskei (TBVC states); KaNgwane, KwaNdebele, KwaZulu, Gazankulu, Qwaqwa and Lebowa (self governing territories). A total of 600 enumeration areas (EAs) were drawn and 10 households were selected from each EA, yielding a sample of about 6 000 households. Sample selection was carried out independently in each stratum applying a two stage sampling procedure; a systematic sample of EAs followed by a systematic sample of households.Former homelands in six provinces were regarded as natural occurring strata and as a result Western Cape, Northern Cape and Gauteng were not part of the strata.
Weighting
Since the sample design was not necessarily a self-weighting design, weights had to be used. Two types of weights were calculated, namely, person weights and household weights. Household weights were calculated by using the reciprocal of the product of an EA inclusion probability and a household inclusion probability. Relative scaling was done on the household weight due to the changes in EA sizes in the 1996 census sample frame and to take into account the growth rate of the population from 1996 to 1997. Person weights were calculated by first post-stratifying the data by province (former homeland in a province) gender and 5 year age groups. The 1996 Census population figures were used as benchmarks for the post stratum. Since the target areas were former homelands, which are largely African areas, population group was a factor in the benchmarks.
