Longitudinal changes in COVID-19 vaccination intent among South African adults: evidence from the NIDS-CRAM panel survey, February to May 2021 – BMC Public Health

Survey design

We analyse data from the latest two waves of the National Income Dynamics Study: Coronavirus Rapid Mobile Survey (NIDS-CRAM), a national and broadly representative longitudinal survey of adults in South Africa, to determine the proportion of adults who reported willingness or hesitancy to receive a COVID-19 vaccine. The objective of the NIDS-CRAM survey was to create a nationally representative, rapid, and longitudinal dataset that could inform evidence-based policy-making during the COVID-19 pandemic. The survey instrument, which is available online, includes a range of questions pertaining to employment, welfare, hunger, government assistance, and COVID-19-related beliefs [7]. The sample was drawn from a subsample of adult respondents from the latest wave of the National Income Dynamics Study (NIDS), a nationally representative longitudinal survey of initially over 28,000 South African adults that tracked social and economic outcomes from 2008 to 2017. Although the original NIDS longitudinal study was administered in person, the NIDS-CRAM study was administered telephonically. Wave 1, which surveyed 7073 adults, was conducted in May and June 2020, shortly after the onset of South Africa’s national lockdown at the end of March. Due to attrition between Waves 1 and 2 (approximately 19%), the sample in Wave 3 was replenished with a top-up sample of 1084 respondents. Our analysis uses data from the latest two waves of NIDS-CRAM that include data on vaccination intent, Waves 4 (conducted in February/March 2021) and 5 (April/May 2021).

Study setting

For context, Fig. 1 illustrates the timing of the five NIDS-CRAM waves with respect to daily new confirmed COVID-19 cases and lockdown alert levels (with level 5 being the most stringent and 1 the most lenient). Important vaccine-related developments that coincided with Waves 4 (February 2 to March 10, 2021) and 5 (April 6 to May 11, 2021) are indicated in Fig. 2 alongside the number of vaccine doses administered. Of note, the vaccination programme was placed on hold twice during this period. On February 7, it was announced that the Oxford-AstraZeneca vaccine had limited efficacy against the dominant Beta variant, and the country’s doses were sold to other African Union member countries [9]. From April 13 to 28, administration of the Johnson & Johnson vaccine was temporarily suspended in light of concerns about its possible association with cerebral venous thrombosis [10]. The total amount of administered vaccinations surpassed 125,000 by the end of Wave 4 and 415,000 by the end of Wave 5 [11]. The surveys also preceded the widespread vaccination of individuals aged 60 years or older, which began on May 17 [12].

Timing of the NIDS-CRAM waves with respect to COVID-19 cases and lockdown levels in South Africa. Authors’ own arrangement. Source of COVID-19 case data: Our World in Data [8]. Solid line represents 7-day rolling average of daily new confirmed COVID-19 cases. L = lockdown level

Full size image

Timeline of NIDS-CRAM survey dates, vaccine-related events, and administered doses in South Africa. Authors’ own arrangement. Source of vaccine dose data: Our World in Data [8]

Full size image

Outcome variables

Vaccine-related survey questions are provided in Additional file 1. Our main outcome variable was COVID-19 vaccination intent. In both waves of the survey, we asked respondents to indicate the extent of their agreement with the statement “if a vaccine for COVID-19 were available, I would get it”, with response options being “strongly agree”, “somewhat agree”, “somewhat disagree”, “strongly disagree”, and “I don’t know”. In Wave 5, respondents were first asked if they had already been vaccinated, and they skipped the rest of the vaccine module if so. Vaccine willingness was defined to include respondents who “strongly” or “somewhat” agreed with the statement, and vaccine hesitancy was defined to include those who “strongly” or “somewhat” disagreed, as well as those who said that they did not know.

To better understand motivations, vaccine-hesitant respondents in Wave 5 were asked whether they thought the vaccine was unsafe or could harm them. If they responded “yes”, they were asked how convinced they were of this, with response options being “a little”, “somewhat”, or “very” convinced. Finally, respondents were asked the open-ended question, “Why do you believe the vaccine is unsafe or harmful?” Interviewers were provided with eight categories (corresponding to findings from exploratory work on vaccine beliefs) for coding responses, but they were instructed not to read out these categories. Responses were coded to existing categories if applicable, or were captured as free text by the interviewer and then later categorised by a research psychologist using thematic analysis.

Covariates

We drew on a wide range of information about respondents’ demographic, ethnic, social, and economic characteristics, collected in the NIDS-CRAM as well as from their records in previous NIDS waves. We included variables capturing settlement type, province, age (18–24, 25–59, 60 and older), gender, population group (black African, Coloured, White, and Asian/Indian), language spoken at home, and self-reported religious affiliation. We used two questions regarding COVID-19 risk beliefs in our analysis: a question asking whether respondents thought they were likely to get the Coronavirus, and a question asking whether they thought they could avoid getting the virus. Regarding medical risk factors, we included biometric data on body mass index and blood pressure from two repeated measurements from NIDS Wave 5 (2017). We also included responses to the question, “Do you have any of these chronic conditions (you don’t have to tell us which one): HIV, TB, lung condition, heart condition or diabetes?” from NIDS-CRAM Wave 1. We also included an open-ended question from NIDS-CRAM Wave 1 asking respondents where they get information about COVID-19 that they trust. Finally, to examine variation in vaccine hesitancy by income or wealth, we relied on several measures of socioeconomic status. Due to concerns about reliability of and bias in a household income variable captured in the survey, we generated a “deprivation and poverty” household asset index as a proxy to capture differences in socioeconomic status (see Additional file 2 for more details). Additionally, we used respondents’ report of recent hunger in the household and receipt of a means-tested state cash transfer (social grant) as proxies for socioeconomic status.

Statistical analysis

For each wave, we conducted cross-sectional analyses on aggregate and between-group variation in vaccine hesitancy. Transition matrices were used to examine individual-level changes in vaccine willingness between NIDS-CRAM Wave 4 and Wave 5. We also employed bivariate descriptive analyses as well as a multivariable linear probability model to examine the correlations between vaccination intent and a large number of demographic characteristics and individual attributes. Estimates were weighted using the relevant sampling weights, drawn from the 2017 NIDS survey, to account for the complex survey design and to adjust for non-random non-response and attrition [13, 14]. In our regression analysis of predictors of vaccine hesitancy and changes in vaccine hesitancy across the two survey waves, we included age, gender, population group, language spoken at home, religious affiliation, beliefs about COVID-19, comorbidities, and trusted information sources. Our analyses employed a 5% significance level to assess the precision of estimates.