Sampling Procedure
CVACS Survey 1 was obtained from a stratified sample drawn from the GeoTerraImage (GTI) 2021 sampling frame (https://geoterraimage.com/), using individuals aged eighteen and older. The sample was primarily stratified across the following categories: province, population group, geographic area type (metro, non-metro urban, non-metro rural) and the neighbourhood lifestyle index (NLI), in groups of NLI 1-2, NLI 3-4, and NLI 5-10. Age categories defined according to the COVID-19 vaccination age groups (18-34, 35-49, 50-59, 60+), and gender were used as further explicit stratification variables. A credit bureau database was linked to this database at the enumeration area level, including individuals who had applied for credit, regardless of the outcome, and individuals who have had a credit check.
The CVACS Sample in Survey 2 included individuals from Survey 1 who were re-interviewed, who fell into two categories: vaccinated between Survey 1 and 2, or those remaining unvaccinated. In order to realise an unvaccinated sample of similar size to Survey 1, a top-up sample of unvaccinated individuals was interviewed. These individuals were drawn from the same sampling frame as Survey 1. Younger and female respondents were less likely to be re-interviewed in Survey 2. The full Survey 2 unvaccinated sample is more skewed to the younger age categories, due to higher vaccination rates among the elderly precluding many from inclusion into the study.
Weighting
The CVACS Survey 2 data includes cross-sectional survey design weights for the full Survey 2 sample of vaccinated and unvaccinated individuals. Panel weights are included for the sample of individuals from Survey 1 who were re-interviewed in Survey 2, whether vaccinated or unvaccinated. Weighted CVACS data, whether design or panel-weighted, is not nationally representative of all unvaccinated individuals in South Africa. Design-weighted CVACS Survey 2 data corrects for non-response and uneven inclusion probabilities within the sampling frame. Panel-weighted CVACS data corrects for a degree of attrition bias. With very high rates of non-response and attrition, the weighting adjustments are unlikely to account for all selection bias in non-response or self-reported vaccination status, or for attrition bias across Survey 1 and 2.
